AW AgentAIWatchPublic threat intelligence
3404 visits
Public sources only

DEFENSIVE INTELLIGENCE

Track suspicious autonomous-agent activity

Every incident includes its source, verification state, confidence score, and an explainable risk assessment.

Total reports364Current filtered view
Confirmed7Officially attributed
High risk3Risk score 70+
Average risk33Out of 100

ACTIVITY

Incident timeline

16 active periods
1
2024-01
1
2024-02
1
2024-03
1
2024-10
1
2025-06
1
2025-07
1
2025-08
2
2025-09
1
2025-12
1
2026-02
1
2026-04
1
2026-06
2
2026-07
4
2026-09
156
2026-10
189
2026-11

GEOGRAPHY

Incident locations

7 located reports
World map of reported incidents
Confirmed Probable Unverified

THREAT RECOGNITION

Attack techniques detected

8 techniques · 8 campaigns
155T1190 Exploit Public-Facing Application
98T1078 Valid Accounts / Credential Abuse
57T1052 Exfiltration
24T1059 Command & Scripting Execution
16T1105 Ingress Tool Transfer
11T1021 Lateral Movement
8TA0042 Jailbreak / Prompt Injection
3TA0043 Agent Misalignment
CVESCVE-2023-46604, CVE-2024-47575, CVE-2025-32711, CVE-2026-101767, CVE-2026-102764, CVE-2026-104935, CVE-2026-106173, CVE-2026-107406, CVE-2026-109618, CVE-2026-109619, CVE-2026-111357, CVE-2026-114938, CVE-2026-118527, CVE-2026-123158, CVE-2026-124194, CVE-2026-125350, CVE-2026-126325, CVE-2026-19868, CVE-2026-35273, CVE-2026-41709, CVE-2026-81648, CVE-2026-83844, CVE-2026-88771, CVE-2026-88772, CVE-2026-88773, CVE-2026-88774, CVE-2026-88775, CVE-2026-88776, CVE-2026-88777, CVE-2026-88778, CVE-2026-88779, CVE-2026-98432
IPS25.1.1.1, 25.1.1.2, 25.1.2.2
DOMAINSagent.datasette.io, claude.ai, datasette.io, divagr.com, simonwillison.net, static.simonwillison.net
TOOLSChatGPT, Claude, Copilot, GPT-5, GPT-6, Gemini, Llama, claude, gemini, gpt-5, gpt-6, llama

ANALYST DESK

Ask the agents

3 agents on duty
Triageidle

Awaiting a question.

Geographeridle

Awaiting a question.

Skepticidle

Awaiting a question.

EVIDENCE LOG

Public incident reports

182026-11
unverified public report Unknown

CISA Urges Immediate Patching of Java4Shell Flaw That Could Allow Remote Code Execution

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has urged organizations to immediately patch a newly disclosed critical security flaw impacting Oracle Java. The flaw, tracked as CVE-2026-114938 (CVSS score: 8.1), refers to a case of out-of-bounds write that can lead to denial-of-service (DoS) or remote code execution under certain conditions. "This critical memory-corruption

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
182026-11
unverified public report Unknown

Attackers Compromise Microsoft's Agent Framework PyPI Package to Steal Cloud Credentials

Threat actors have compromised Microsoft's agent-framework package to push a credential stealer, according to reports from Aikido, Chainguard, Endor Labs, OX Security, SafeDep, Socket, StepSecurity, and Google-owned Wiz. The malicious versions, 1.26.1 and 1.26.2, were published 33 minutes apart on November 17, 2026, and are designed to run a credential stealer on every Python start. As of

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
182026-11
unverified public report Unknown

Russian COLD RELIC Hackers Use Stolen Google Workspace Data to Breach Second U.S. Organization

The Russia-linked cyber espionage group known as Cold Relic (formerly COLDRIVER and UNC4057) has been observed weaponizing a single cloud compromise into a launchpad for a series of linked intrusions resulting in third-party compromise. Google's Mandiant and Threat Intelligence Group (GTIG) said it investigated a 2026 compromise at a U.S.-based organization in which the adversary leveraged

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
182026-11
unverified public report Unknown

Uber says nearly 150,000 impacted by third-party breach

In breach notification letters, the rideshare company said it was notified by a third party service provider on October 9 that internal credentials were stolen and used to access Uber’s account.

Risk rationale: credential (+12) therecord.media ↗
30Risk
24Confidence
172026-11
unverified public report Unknown

External sandboxes for coding agents

External sandboxes for coding agents I've been exploring new ways to sandbox code that is executed by agents, with a focus on making it safer to try out local models without worrying that a mistake or prompt injection attack might cause them to take damaging actions on a machine. I worked on this project with Prime Radiant , the AI research lab behind Superpowers and other projects . My post on the Prime Radiant blog introduces the external sandbox pattern, where code executed by the agent runs inside a dedicated sandbox while the agent itself lives outside of that sandbox. To demonstrate this pattern we've released three sandbox server implementations: external-sandbox-v86 runs a Node.js server which then uses v86 to run Alpine Linux inside WebAssembly. external-sandbox-container is a Python web server which runs the sandbox in Alpine Linux inside an Apple Containers instance. external-sandbox-quicksand is another Python web server, this time using Microsoft's Quicksand sandboxing lib

Risk rationale: model (+5), prompt injection (+20) simonwillison.net ↗
43Risk
24Confidence
172026-11
unverified public report Unknown

GPT‑6.2 Astra and GPT‑6.2 Sol

GPT‑6.2 Astra and GPT‑6.2 Sol OpenAI: We’re introducing GPT‑6.2 Astra, our new frontier model for your hardest projects, and GPT‑6.2 Sol, our efficient, versatile model for everyday work. They're available in the API as gpt-6.2-astra and gpt-6.2-sol . I shipped LLM 0.38 with support for the new models. 6.2 Astra is 80% of the price of 6 Astra - $8/MTok input and $40/MTok output, the same price as Claude Fable 5.2 (which was also 80% of its predecessor). 6 Astra and Fable 5 were both $10/$50. The cached price is an even bigger discount - $0.40/million compared to $1/million for the older model. This is great news for longer running agentic sessions. GPT-6.2 Sol is the same $2/$10 price as 6.1 Sol and 6 Sol - which are half the price of GPT-5.6 Sol ($4/$20). Here are the pelicans for GPT-6.2 Sol and for GPT-6.2 Astra . The 6.2 Astra max pelican looks like this: I have a full comparison grid here , but here's how the high/xhigh/max pelicans compare with the previous Sol and Astra releases

Risk rationale: agentic (+14), model (+5) simonwillison.net ↗
37Risk
24Confidence
172026-11
unverified public report Unknown

Chinese state group developing automated attack tool for cyber espionage, Google says

Google released its latest AI Threat Tracker report on Monday, writing that threat actors are rapidly evolving their use of artificial intelligence in attacks — now leaning more heavily on autonomous multi-agent offensive frameworks like Artex, CAIRN, CyberStrikeAI and PentAGI to execute multi-stage intrusions at greater operational velocity.

Risk rationale: intrusion (+14) therecord.media ↗
32Risk
24Confidence
172026-11
unverified public report Unknown

Oracle to Patch 11 Java Flaws Today as Critical OpenJDK RCE (CVE-2026-114938) Report Emerges

Oracle is preparing to release security patches for 11 vulnerabilities in Java SE on Tuesday, November 17, 2026, all of which may be exploited remotely without authentication, as separate reports warn of a potentially critical remote code execution (RCE) flaw affecting OpenJDK. The 11 vulnerabilities carry a maximum CVSS 3.1 severity score of 8.1, according to Oracle's advance security notice.

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
172026-11
unverified public report Unknown

Formal Verification of the Signal Protocol Using AI

This seems like a good thing: Google Deepmind researchers used agentic AI to mathematically prove, or “formally verify,” a number of the Signal protocol’s key cryptographic guarantees­a landmark development for the widely used cryptography and AI itself. Conducted as a sort of proof-of-concept project in just four weeks, the work is significant in two ways. It serves as further independent validation of the cryptography that protects massive quantities of the world’s communications on platforms like the Signal app, Google Messages, and WhatsApp. And it demonstrates Google DeepMind’s formal verification capabilities using a new AI framework called AlphaProof Code...

Risk rationale: agentic (+14) schneier.com ↗
32Risk
24Confidence
172026-11
unverified public report Unknown

Critical Java Vulnerability Requires Immediate Patching

Purpose CISA strongly urges organizations to immediately patch the recently disclosed Oracle Java vulnerability CVE-2026-114938 . This critical memory-corruption vulnerability, known as “Java4Shell,” affects the Base64 decoder of the x86-64 HotSpot Java Virtual Machine (JVM), which is part of the Java Development Kit (JDK). Specially crafted Base64 input could crash the JVM and allow for unauthenticated remote code execution. Actions to take CISA urges organizations to take the following actions: Identify affected Java installations. Review your software inventory for JDKs and Java runtimes, including those bundled with third-party applications or included in container images and appliances. Note: JDKs and Java runtimes may not appear as separate entries in software inventories. Apply relevant Java and vendor patches as soon as they are available, prioritizing internet-facing systems. If immediate patching is not feasible, implement temporary compensating controls, such as restricting

Risk rationale: remote code execution (+18) cisa.gov ↗
36Risk
39Confidence
172026-11
unverified public report Unknown

New GitDoor and GraphDoor Backdoors Turn GitHub and SharePoint Into Covert C2 Channels

Government sectors in Asia and Africa have been targeted by an active cyber intrusion campaign that delivers to previously undocumented backdoors that leverage legitimate services like Microsoft SharePoint and GitHub for command‑and‑control (C2). The activity has been codenamed Operation ShadowCloud by Kaspersky. The Russian cybersecurity company said it identified the campaign in early 2026.

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
172026-11
unverified public report Unknown

Apple Tightens macOS App Permissions to Limit Data Access Amid AI Agent Security Risks

Apple on Monday introduced new controls that allow Mac users to give more granular access to data they give to apps, including those that use artificial intelligence (AI) agents. The changes, currently previewed in macOS 27.3 beta 2, relate to increased transparency and more options for two of its APIs: Device Control and Data Access, and Full Disk Access. "Before granting an app such

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
162026-11
unverified public report Unknown

California sues OpenAI over AI agents escaping sandboxes, hacking firms

California Attorney General Rob Bonta sued OpenAI on Monday, alleging the company knowingly continued testing advanced artificial intelligence models despite repeated security failures that allowed autonomous AI agents to escape isolated environments and launch cyberattacks against real-world organizations. [...]

Risk rationale: ai agent (+18), model (+5) bleepingcomputer.com ↗
41Risk
24Confidence
162026-11
unverified public report Unknown

Cursor Supply Chain Attack Started at Fireworks AI

The attacker stole an access key from Fireworks and used it to exfiltrate credentials that enabled them to push malicious Claude Code extension copies. The post Cursor Supply Chain Attack Started at Fireworks AI appeared first on SecurityWeek .

Risk rationale: exfiltrat (+16), credential (+12) securityweek.com ↗
46Risk
24Confidence
162026-11
unverified public report Unknown

Barracuda Hacked Amid Surge in Artex-Powered Cyberattacks

Chinese threat actors are broadly adopting the agentic AI penetration testing framework to accelerate their attacks. The post Barracuda Hacked Amid Surge in Artex-Powered Cyberattacks appeared first on SecurityWeek .

Risk rationale: agentic (+14) securityweek.com ↗
32Risk
24Confidence
162026-11
unverified public report Unknown

Google Finds Hackers Using AI Agents to Build and Run Exploits From Software Patches

Google on Monday revealed that adversarial use of artificial intelligence (AI) by threat actors has gone beyond manual prompting to deploying autonomous multi-agent offensive frameworks like ARTEX, CAIRN, CyberStrikeAI, and PentAGI to execute multi-stage intrusions at greater operational velocity. In tandem, threat actors are leveraging the power and capabilities of large language models (LLMs)

Risk rationale: ai agent (+18), model (+5), intrusion (+14) feeds.feedburner.com ↗
55Risk
24Confidence
162026-11
unverified public report Unknown

SANSAR Linux Cryptomining Malware Uses Modified NVIDIA Library to Hide GPU Activity

A financially motivated threat activity cluster tracked as REF3364 has been attributed to a Linux-focused cryptocurrency mining operation that tampers with the readings used to detect high GPU workloads. Central to the intrusion is a toolkit called SANSAR, a Rust-based orchestrator that receives commands over the TOR anonymity network to manage CPU and GPU miners. The activity was observed in

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
162026-11
unverified public report Unknown

Amazon Uncovers AI Agent Campaign Targeting LiteLLM, Dify, and Other Apps for API Keys

Attackers' AI agents have been hunting for weak spots in internet-exposed AI apps such as LiteLLM, One API, and Dify to steal the AI provider keys stored inside, according to the agents' own records. The records, which Amazon's threat intelligence team captured with decoy servers, describe stolen keys being used to run the AI models that powered the search for more keys. The team said this

Risk rationale: ai agent (+18), model (+5) feeds.feedburner.com ↗
41Risk
24Confidence
142026-11
unverified public report Unknown

datasette-llm 0.1a11

Release: datasette-llm 0.1a11 datasette-llm is the intermediate plugin for Datasette that supports configuring LLMs for use with other Datasette plugins. This release adds support for OpenAI-compatible endpoints, so you can configure Datasette to use an LLM hosted anywhere that talks the OpenAI pseudo-standard web protocol. This means that if you're running local models with LM Studio or Ollama or llama.cpp (which just hit v1.0.0 ) you can run Datasette via uvx like this: wget https://datasette.io/content.db uvx --with datasette-agent datasette \ --root --secret 1 \ -s plugins.datasette-llm.default_model ' { "model": "lmstudio-qwen", "model_name": "qwen/qwen3.8-27b", "api_base": "http://localhost:1234/v1" } ' content.db And you'll get a local server running a Datasette Agent instance powered by your local model! Tags: uv , datasette-agent , llm , datasette , llms , ai , generative-ai

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
132026-11
unverified public report Unknown

In Other News: New CISA Director Nominee, Connected Car Hacking, Baker Tilly Data Breach

Noteworthy stories that might have slipped under the radar: Siemens details security advisory policy, OpenAI Codex backdoor risk, Microsoft warns of model weight backdoors. The post In Other News: New CISA Director Nominee, Connected Car Hacking, Baker Tilly Data Breach appeared first on SecurityWeek .

Risk rationale: model (+5) securityweek.com ↗
23Risk
24Confidence
132026-11
unverified public report Unknown

How one AI agent exposed five critical enterprise security gaps

An AI agent built to automate invoice processing could be manipulated into changing vendor bank details through a malicious email, without triggering traditional security alerts. Token Security examines five critical AI agent security gaps and shares practical guidance from industry security leaders on preventing them. [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
132026-11
unverified public report Unknown

When an AI Agent Goes Wrong, Can You Stop It?

Suppose an AI agent is asked to sort support tickets and look up account details. A ticket contains instructions that lead the agent to export records to an unfamiliar destination. The agent can make the request because its identity has access to the records and its connected tools allow the action. That kind of scenario raises a response question: what does “stop the agent” actually mean? A

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
132026-11
unverified public report Unknown

OpenAI Unveils Daybreak Red Team, Defense Factory Security Tools

Daybreak Red Team pentests web applications with AI agents, while Defense Factory investigates security findings and prepares patches. The post OpenAI Unveils Daybreak Red Team, Defense Factory Security Tools appeared first on SecurityWeek .

Risk rationale: ai agent (+18) securityweek.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

llm-gemini 0.37

Release: llm-gemini 0.37 Google launched Gemini 4 Hydrogen today - the lowest cost member of the Gemini 4 family. It's actually not too far off the price of its much bigger siblings, Carbon and Argon. I'm finding Gemini 4's pricing really confusing. Here's the pricing scheme: Model Dates Input ($/M) Cached input Output Gemini 4 Carbon Through Nov 30, 2026 $2.00 $0.10 $10.00 From Dec 1, 2026 $4.00 $0.20 $20.00 Gemini 4 Argon Through Nov 30, 2026 $2.00 $0.10 $10.00 From Dec 1, 2026 $4.00 $0.20 $20.00 Gemini 4 Hydrogen   $1.50 $0.075 $7.50 Carbon and Argon are the same price , and both are scheduled to double in price on December 1st. If they're the same price, why have two different models? Google says that Argon "delivers Gemini 4 intelligence with faster responses and lower token spend per turn compared to Gemini 4 Carbon" - so it may be priced the same, but it will use less tokens per task. Hydrogen, supposedly the cheap model, is currently only a 25% discount on its more capable

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
122026-11
unverified public report Unknown

New N-able "4EverAble2Shell" zero-day exploited hours after PoC released

Hackers are actively exploiting a newly disclosed N-able N-central zero-day vulnerability dubbed "4EverAble2Shell" to compromise servers and deploy cryptocurrency miners, with researchers reporting at least 21 affected systems, including 19 running the company's latest security update. [...]

Risk rationale: zero-day (+18) bleepingcomputer.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

Webinar: See How Identity Blind Spots Put Your SOC at Risk and How to Close Them

Security teams have more alerts, logs, and monitoring tools than ever. Yet when an attack happens, one simple question can still take too long to answer: Who or what is behind this activity, and what can they access? A suspicious login might belong to an employee. An unusual API call could come from a machine account. Or an AI agent might be accessing sensitive systems using permissions nobody

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

New SignDrop Zero-Day Exploit Turns Windows CLFS Security Feature Into SYSTEM Access

Public exploit code released earlier this week claims that a local Windows user can gain SYSTEM privileges through the Common Log File System driver. The proof of concept, called SignDrop, targets a log-authentication feature added to the driver last year. It abuses how clfs.sys creates companion files when a privileged process opens a prepared log, using path redirection to take control of

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

PTC Patches Critical Windchill and FlexPLM Flaw That Could Enable Remote Code Execution

PTC has released patches for a new critical vulnerability impacting Windchill and FlexPLM software that, if successfully exploited, could result in remote code execution. The security flaw, tracked as CVE-2026-126325, carries a CVSS v4 score of 9.2 out of a maximum of 10.0. It has been described as a case of improper input validation and deserialization of untrusted data that could result in

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

Inside an AI Attack Against a US Retailer

When hand-typed commands failed to breach a US retailer, the attacker turned to AI-generated scripts, achieving lateral movement in minutes.

Risk rationale: lateral movement (+16) darkreading.com ↗
34Risk
24Confidence
122026-11
unverified public report Unknown

Hackers Target N-central Zero-Days as N-able Rushes Patches

The malicious activity started shortly after a researcher dropped over 30 zero-days. N-able patched the unauthenticated ones. The post Hackers Target N-central Zero-Days as N-able Rushes Patches appeared first on SecurityWeek .

Risk rationale: zero-day (+18) securityweek.com ↗
36Risk
24Confidence
122026-11
unverified public report Unknown

Attacker Used 4,837 Hijacked GitHub Accounts to Run a CI Secret Stealer in 1,806+ Repositories

Starting Tuesday, an attacker used stolen credentials for 4,837 GitHub accounts to plant a secret stealer in repositories those accounts could write to, according to Aikido Security, a security software company. The stealer is built to take the secrets and cloud credentials that a repository's CI workflows can reach. It ran in at least 1,806 repositories, where the attacker has likely collected

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
112026-11
unverified public report Unknown

Gemini Decisions API

Gemini Decisions API Google join Jev and OpenAI in the decision model category with "Gemini Decision 1". This new model API is almost identical to Jev: it supports the same three types of questions (yes/no, choice, and rubric-scoring) and even reuses Jev's term "Noul" for a yes/no question. It's priced the same as the GPT-6 Luna decision model: 10 cents per million input tokens, and output is free. Jev remains the cheapest of the three, at 4.2 cents per million. Jev is text (and JSON) only. OpenAI can handle text and images. Gemini Decision 1 is fully multi-model: it covers text, images, audio, video, and PDFs. I tested it against video by passing these questions to it along with my Yonkers stomping robot video file: dragon: Does this video have a dragon in it? robot: Does this video have a robot in it? france: Is this video set in France? usa: Is this video set in the USA? Jev and OpenAI both return 1.0 or 0.0 for yes/no questions, but Gemini insists on being a little less confident t

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
112026-11
unverified public report Unknown

aws-emulator

aws-emulator Finally, an official emulator from AWS to run tests against a local clone of their many different services without having to provision a paid API key to do so. It's written in Rust by a huge swarm of agents - 29,684 commits in the first three weeks of the project. Swaminathan (Swami) Sivasubramanian wrote about how it came to be in a guest post on Werner Vogels' blog. It currently requires Docker to run on macOS or Windows. I'm hoping they fix that in the future - it would be great if you could download a single binary and use that as part of a standard test suite in other languages and on other platforms. Tags: amazon , amazon-web-services , testing , werner-vogels , ai , rust , docker , generative-ai , llms , agentic-engineering

Risk rationale: agentic (+14) simonwillison.net ↗
32Risk
24Confidence
112026-11
unverified public report Unknown

llm 0.37.2

Release: llm 0.37.2 Fixes a security issue reported by Daniel de Araujo in GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w . LLM has a mechanism called extra-openai-models.yaml - now managed by the new-in-0.37 llm openai endpoint add command - for using LLM with OpenAI-compatible API endpoints from other providers. Daniel reported two ways this could accidentally send your configured OpenAI API key to one of those providers: if the key named by api_key_name had never been stored, or if a responses: true model was run with -o chat_completions 1 . Tags: projects , security , llm , releases

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
112026-11
unverified public report Unknown

AntiFox AI Pentesting Platform Linked to Attacks That Accessed AWS Production Secrets

Cybersecurity researchers have disclosed details of a campaign in which an emerging agentic penetration testing platform called AntiFox has been used to steal Amazon Web Services (AWS) credentials and pivot laterally into compromised environments. "In several recently observed intrusions, Alibaba Cloud systems hosting AntiFox interfaces were associated with confirmed malicious actions against

Risk rationale: agentic (+14), credential (+12), intrusion (+14) feeds.feedburner.com ↗
58Risk
24Confidence
112026-11
unverified public report Unknown

Anthropic's IPO roadshow video

Anthropic's IPO roadshow video Anthropic kicked off their IPO roadshow yesterday, and their roadshow video has some previously unreleased information about how their business works. The section on "Approach to compute" with co-founder Tom Brown is particularly interesting, describing how Anthropic train and serve Claude across AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs. I'm not sure why there's no official video transcript, but here's one that I extracted using MacWhisper. Highlights below: We train and serve Claude now across all four major chip families. That's only possible because early on we made an unconventional bet on a multi-chip strategy. At the time, many people thought that running AI models on different chips wasn't worth the effort. Each chip needs its own tailored software stack and inference stack, which is expensive and difficult to build. We've now been doing it for years, and we've gotten incredibly good at it. [...] As an example, this year, we starte

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
112026-11
unverified public report Unknown

Blind trust: stealing credentials through CSP

Content Security Policy is supposed to be the safety net that stops an HTML injection from turning into XSS. Usually it works. But sometimes it doesn't, because the policy allows-lists a domain that h

Risk rationale: credential (+12) portswigger.net ↗
30Risk
24Confidence
112026-11
unverified public report Unknown

Citrix, Ivanti Patch Critical and High-Severity Vulnerabilities

Attackers could exploit the bugs to tamper with certificate entries, achieve remote code execution, and compromise hosts. The post Citrix, Ivanti Patch Critical and High-Severity Vulnerabilities appeared first on SecurityWeek .

Risk rationale: remote code execution (+18) securityweek.com ↗
36Risk
24Confidence
112026-11
unverified public report Unknown

Fortinet Patches Nearly 100 Vulnerabilities

The massive drop comes after the company integrated frontier AI models into its product security testing. The post Fortinet Patches Nearly 100 Vulnerabilities appeared first on SecurityWeek .

Risk rationale: model (+5) securityweek.com ↗
23Risk
24Confidence
112026-11
unverified public report Unknown

We Built An Agent That Broke All of Our Security Rules. Here's What We Learned.

We tell customers three things about securing AI agents: they need their own identities, access that fits the job they’re given, and traceability that’s separate from any human’s. That was true for us at Token as well. Then, we built an agent that broke all three by default. This helped us think through how such agents should be defended. When a browser agent acts inside a user’s authenticated

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
112026-11
unverified public report Unknown

Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons

An AI model can be given a hidden backdoor by editing fewer than 10 of its rarely used neurons, without retraining the model, Microsoft researchers have shown. When a chosen trigger appears, the model outputs an attacker's command, which an agent using that model would run with the agent's own permissions. The attacker must be able to change the model's weights. The team tested the method on

Risk rationale: ai agent (+18), model (+5) feeds.feedburner.com ↗
41Risk
24Confidence
112026-11
unverified public report Unknown

Microsoft Patch Tuesday for November 2026 — Snort rules and prominent vulnerabilities

Microsoft has released its monthly security update for November 2026, which includes 1328 vulnerabilities affecting a range of products, including 60 that Microsoft marked as "critical." Out of 60 "critical" vulnerabilities, 45 are remote code execution (RCE) vulnerabilities.

Risk rationale: remote code execution (+18) blog.talosintelligence.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

Railway breach leads to $740,000 in fraudulent AI charges

Cloud hosting provider Railway says hackers exploited vulnerabilities in its infrastructure to steal customer authentication secrets and generate approximately $740,000 in fraudulent AI usage, potentially exposing credentials for up to 100,000 workspaces. [...]

Risk rationale: credential (+12) bleepingcomputer.com ↗
30Risk
24Confidence
102026-11
unverified public report Unknown

Microsoft Fixes Nearly 1,100 Security Vulnerabilities

Microsoft did another four-digit patch drop today, releasing updates to fix almost 1,100 security vulnerabilities in its Windows operating systems and supported software. This month's Patch Tuesday addresses 63 critical flaws -- including a zero-day in Microsoft Outlook -- and comes amid important security updates from other software makers, including Adobe, Apple, Google and Mozilla.

Risk rationale: zero-day (+18) krebsonsecurity.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

llm-gemini 0.36

Release: llm-gemini 0.36 Gemini 4 Carbon was announced two weeks ago for "a subset of cyber defenders", but today it's available to everyone else . This is Google's best ever Gemini, priced at $2/million input and $10/million output at launch (that's GPT-6.1 Sol / Claude Sonnet 5.5 prices) but scheduled to double in price to $4/$20 on December 1st (same as Opus 5.5). Here are three pelicans for the low, medium and high thinking efforts (it defaults to medium). This is the pelican I got for high: Gemini pelicans have all been great for a while now, so I also tested it out using Google's Antigravity IDE with a prompt I first tried with Claude Fable 5.2 yesterday : Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and stomps around manhattan and queens and Brooklyn, use openstreetmap mapping data Here's the HTML result (I had to prompt "build that into a standalone HTML file" to get a version with no external dependencies). I then screen-captured that

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
102026-11
unverified public report Unknown

SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users

An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and ran code on one of NEAR AI's inference servers last Sunday. The attacker could also have seen requests being processed on that server that day, including requests sent in confidential mode, the company said earlier this week. That mode processes requests inside a trusted execution environment (TEE

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

Adobe Patches 110 Vulnerabilities Across 9 Products

Adobe has fixed vulnerabilities in AEM, Premiere, Photoshop, After Effects, Connect, Audition, Dimension, Content Credentials SDK, and Media Encoder. The post Adobe Patches 110 Vulnerabilities Across 9 Products appeared first on SecurityWeek .

Risk rationale: credential (+12) securityweek.com ↗
30Risk
24Confidence
102026-11
unverified public report Unknown

Fireworks AI Breach Widens to Expose Customer Billing Data and Incomplete Model Weights

Fireworks AI disclosed on Monday that its investigation into a security incident first reported early last month  uncovered further malicious activity, affecting account and billing information, alongside secrets, datasets, and incomplete models for a "limited set of accounts." In a notice published on October 10, 2026, the company said it detected unauthorized use of internal Fireworks

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
102026-11
unverified public report Unknown

Fireworks AI Confirms Wider Breach Affecting Customer Datasets, Model Weights and Notion AI Chats

Fireworks AI disclosed on Monday that its investigation into a security incident first reported early last month  uncovered further malicious activity, affecting account and billing information, alongside secrets, datasets, and incomplete models for a "limited set of accounts." In a notice published on October 10, 2026, the company said it detected unauthorized use of internal Fireworks

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
102026-11
unverified public report Unknown

WhiteMatter Raises $20 Million in Series A Funding

The cybersecurity startup deploys a dedicated guardian agent for every AI agent in an organization’s environment. The post WhiteMatter Raises $20 Million in Series A Funding appeared first on SecurityWeek .

Risk rationale: ai agent (+18) securityweek.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

Over a Dozen Organizations Hit in Qubole Supply Chain Attack

Hackers used compromised Qubole IAM credentials to steal customer files, secrets, keys, environment variables, and credentials. The post Over a Dozen Organizations Hit in Qubole Supply Chain Attack appeared first on SecurityWeek .

Risk rationale: credential (+12) securityweek.com ↗
30Risk
24Confidence
102026-11
unverified public report Unknown

Binance Warns of iOS Sandbox Escape That Could Expose Crypto Wallet Keys, No Fix Confirmed

An iOS sandbox escape published on GitHub lets an app read other apps' files on iPhones running iOS 26 and 27. That puts seed phrases, private keys, and login credentials at risk of exposure to malicious apps, Binance warned today. Apple's updates for iOS 26 and 27, released yesterday, do not list a fix. A researcher who goes by forcequit,  in collaboration with 0xjohnny, 

Risk rationale: sandbox escape (+24), credential (+12) feeds.feedburner.com ↗
54Risk
24Confidence
102026-11
unverified public report Unknown

SAP Patches Four CVSS 10.0 Flaws, Including Two Enabling Remote Code Execution

Four flaws that attackers could exploit without logging in were fixed today as part of SAP's monthly Security Patch Day—the vendor rates all four 10.0, the maximum CVSS score. Three are in the SAP NetWeaver application server, the software SAP systems run on, and one is in the Web Dispatcher, which receives web requests from the internet and passes them on to that server. The flaws in the Web

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

N-able Takes N-central Offline as Researcher Drops Over 30 Zero-Days

Attackers could exploit the flaws for remote code execution, credential theft, and administrator account creation. The post N-able Takes N-central Offline as Researcher Drops Over 30 Zero-Days appeared first on SecurityWeek .

Risk rationale: credential (+12), zero-day (+18), remote code execution (+18) securityweek.com ↗
66Risk
24Confidence
102026-11
unverified public report Unknown

Anthropic's Claude Fable 5.2 Beats GPT-6 Cyber in Multi-Stage Cyberattack Benchmark

Anthropic on Monday released Claude Fable 5.2 and Claude Mythos 5.2 models, which it said exhibit improvements in coding, reasoning, and vision. During pre-release training, the large language models (LLMs) helped its engineers "complete otherwise-infeasible software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
102026-11
unverified public report Unknown

Introducing Our Smallest, Most Efficient Model: Muse Ember

Introducing Our Smallest, Most Efficient Model: Muse Ember Meta AI just launched their entry in the Claude Haiku/GPT Luna category: Muse Ember, which they describe as: [...] our smallest and most efficient model designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. For shorter prompts it's priced the same as GPT-6 Luna and Claude Haiku 5.5 - both are $0.10/million input tokens and $0.50/million output. Today's other big model release, Claude Fable 5.2, is 80x that price at $8/$40. The pricing is a little less straightforward as prompts get longer, though. Luna, Ember, and Haiku all support up to a million input tokens, but they also all have a threshold after which the price goes up. Taking that threshold into account Ember is priced between the two others: Model Prompt size Input ($/M) Cached input Output GPT-6 Luna ≤272k $0.10 $0.01 $0.50 >272k $0.20 $0.02 $0.75 Muse Ember 1.0 ≤200k

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
102026-11
unverified public report Unknown

Google Issues Chrome Security Update for Eighth Actively Exploited Zero-Day of 2026

Google on Monday released updates to patch nine security vulnerabilities, including one that has come under active exploitation in the wild. The high-severity vulnerability, assigned the CVE identifier CVE-2026-125350 (CVSS score: 8.8), has been described as a use-after-free bug in the XML component. "Use-after-free in XML in Google Chrome prior to 157.0.8103.50 allowed a remote attacker

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

N-able Takes N-central Cloud Offline After Researcher Discloses 33 Zero-Days with PoC Exploits

N-able said it has taken N-central Hosted Environments offline and is urging on-premises customers to shut down their servers until further notice following the public disclosure of over 30 vulnerabilities collectively tracked as Able2Shell. "As a precautionary measure, we are temporarily taking our Hosted N-central environment offline while we take the necessary actions to secure the

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

Chrome Update Patches Exploited XML Zero-Day, Eighth This Year

The high-severity use-after-free in Chrome’s XML component was reported anonymously on November 6 and is already exploited in attacks. The post Chrome Update Patches Exploited XML Zero-Day, Eighth This Year appeared first on SecurityWeek .

Risk rationale: zero-day (+18) securityweek.com ↗
36Risk
24Confidence
102026-11
unverified public report Unknown

Quoting Prompting Claude Fable 5.2

Claude Fable 5.2 pays close attention to who owns the code it works on. In Anthropic’s testing on small coding tasks, the model tended to add to the existing structure when the prompt said nothing about ownership or presented the code base as someone else’s, even when that duplicated existing code. When the prompt said the code base was the model’s own, it often consolidated duplicated code and sometimes made related fixes without being asked. If you want the model to improve the structure as it works, say that the code is its own: You are the owner and long-term maintainer of this codebase, and its design is up to you. The ownership sentence makes the model much more likely to consolidate duplicated code into one shared function. In Anthropic’s testing on small coding tasks, the model consolidated duplicated code in under 10 percent of runs without the sentence and in about 75 percent of runs with it. — Prompting Claude Fable 5.2 , Anthropic documentation Tags: prompt-engineerin

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
102026-11
unverified public report Unknown

Yonkers Rises

Tool: Yonkers Rises I realized I didn't know where Yonkers was, so I looked it up on Google Maps and then, to help embed its location and shape in my head (and to try out Claude Fable 5.2's 3D modeling and animation skills), I prompted Fable 5.2 Medium to: Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and átomos around manhattan and queens and Brooklyn, use openstreetmap mapping data It correctly deduced that "átomos" was a mobile keyboard typo for "stomps": Tags: three-js , llms , ai , generative-ai , yonkers-rising , claude-mythos-fable

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
092026-11
unverified public report Unknown

llm-anthropic 0.31

Release: llm-anthropic 0.31 Anthropic released Fable 5.2 and Mythos 5.2 today. Fable 5.2 is currently ranked first on Artificial Analysis . Anthropic say it "solved two open problems in the physics of black holes", so clearly their internal benchmarks have become more ambitious! I've added the new models to llm-anthropic , so you can now run prompts through Fable 5.2 like this: uv tool install llm llm install llm-anthropic llm keys set anthropic # paste key here llm -m claude-fable-5.2 "Generate an SVG of a pelican riding a bicycle" I ran pelicans at all five reasoning levels - low , medium , high , xhigh , and max . You can see them all here - here's the best result I got, from xhigh : max failed to return an image at all, running out of tokens after spending 128,000 thinking about them . That's the same bug that I saw with Opus 5.5 - but it's new for the Fable series, both Fable 5 and Fable 5.1 returned max effort pelicans just fine. It's priced at $8/million input and $40/million ou

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
092026-11
unverified public report Unknown

N-able takes cloud platform offline after 30+ "zero-days" published

N-able has taken its hosted N-central environments offline and urged on-premises customers to shut down their servers after a security researcher publicly disclosed more than 30 alleged zero-day vulnerabilities, including exploits that could let attackers take over vulnerable systems. [...]

Risk rationale: zero-day (+18) bleepingcomputer.com ↗
36Risk
24Confidence
092026-11
unverified public report Unknown

Hackers Exploit Qubole's Trusted AWS Access to Breach Customer Accounts and Steal Data

An attacker broke into AWS accounts belonging to multiple Qubole customers between October 14 and October 26 and stole stored secrets and data, cloud security company Wiz said today. The attacker used credentials for an AWS account that Qubole's platform uses to reach those customers. How the attacker got those credentials has not been disclosed. Qubole, a data lake platform owned by Idera,

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
092026-11
unverified public report Unknown

The Biggest Threat to Your GRC Job Isn’t AI. It’s Ignoring AI.

AI agents are taking over routine governance, risk, and compliance tasks, forcing GRC professionals to rethink the skills needed to stay relevant. Anecdotes examines three essential skills for the AI-driven GRC workplace: data literacy, agent governance, and communication. [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
092026-11
unverified public report Unknown

⚡ Weekly Recap: SSHole Exploited, APT Root RCE, Robot Takeovers, 15K Hacked Sites, and More

A system doesn't always need a complicated bug to let someone in. Sometimes, a blank password is enough. That's one of the bigger stories this week. And it gets worse when the tools meant to keep systems updated or running safely become part of the attack. Elsewhere, attackers are hiding inside normal web traffic, stealing credentials from network devices, and finding new uses for exposed

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
092026-11
unverified public report Unknown

Taiwan-Based Hacker Used BOSSAI AI Agents to Steal Taipei Emergency Dispatch Records

A single attacker using AI agents pulled emergency dispatch records covering seven Taipei districts, researchers at Acronis's Threat Research Unit said last week. The agents also found a Taipei fire department platform exposing patient data from 2018, including ECG readings and ambulance video footage, without a login. The researchers are highly confident that the attacker is a single person

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
092026-11
unverified public report Unknown

The Web Is More Than Enough, by Paul Ford

My comment on The Web Is More Than Enough, by Paul Ford — Hacker News. OK, this comment convinced me to give that a shot. Here's a thing I made in Three.js this morning - a Nudibranch themed Faberge Egg: https://static.simonwillison.net/static/2026/Hidden-Reef-Egg... (77KB) I just asked Claude (Opus 5.5 Medium running in regular Claude Chat) to port that to Vanilla JS, in the same thread where it had built it, and got this: https://static.simonwillison.net/static/2026/Hidden-Reef-Egg... (98KB) They aren't identical - you can see what Claude said about that in the transcript (I don't fully understand why it differed): https://claude.ai/share/e56d163e-2793-4677-aa11-191d5058b414 But yeah, it turns out Claude Opus 5.5 is perfectly capable of 3D modeling in Vanilla JS + WebGL without Three.js. In eyeballing the code I find the Three.js slightly easier to understand myself, but I rarely look at the code for these 3D projects so that isn't a particularly big deal. Tags: claude , javasc

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
092026-11
unverified public report Unknown

Hide your slop PR descriptions behind

Most of the changes I make to my projects these days come in the form of PRs opened for me by Claude Code or Codex. The PR is my unit of work - I use them to bundle commits together, iterating on them until a feature is complete, document decisions and then merge the finished result. Today's coding agents are great at writing PR descriptions. They're too good at describing the work The problem is that they are too good. An agent-written PR description is accurate and comprehensive. It describes the problem, the detailed solution, and the steps taken to verify it. That's a lot of information to absorb. It's not necessarily slop, but it reads like slop! It's very easy to spot an AI-written PR, and as with other forms of AI writing my instant reaction is best summarized as "I'm not reading all of that". It doesn't matter how accurate or useful the information is, I'm instantly put off by the thought that it will take me longer to read than it took the driving author to write... even when

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
062026-11
unverified public report Unknown

OpenAI Launches GPT-6 Cyber After AI Helps Find 81 Firefox Flaws, 948 Linux Exploits

OpenAI on Thursday unveiled GPT‑6 Cyber, which it described as a major step forward in security research that brings together stronger cybersecurity capabilities with a step-change improvement in model alignment. The new model is "more capable at finding and fixing vulnerabilities, and far more reliable staying within the boundaries researchers set," OpenAI said. It has classified the model as

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
062026-11
unverified public report Unknown

Microsoft Paint RCE Fix Now Available, but Windows Update Won't Deliver It

A fix for CVE-2026-83844, a remote code execution flaw in Windows Paint, is now in the Microsoft Store. Paint gets this fix only through a Store update, not through Windows updates. The fixed build, 11.2609.401.0, was the newest Paint build in all ten countries The Hacker News checked today. The flaw is an out-of-bounds read that could let an attacker run code over a network. Microsoft rates

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
062026-11
unverified public report Unknown

The Flood is Coming And The Pipes Were Already Full

An update on Athena, and some unsolicited predictions on what's coming next On Saturday, Dario Amodei published We Must Pace the Frontier. I agree with him. Go read it if you haven't. I've spent the summer looking at what these models find, and the gap he describes between what they can do and what the rest of us are ready for is the one I've been staring at since May. The hardest piece of his

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
062026-11
unverified public report Unknown

Chainalysis Breach Exposes All Users' Credentials, Forces Shutdown of Key Services

Blockchain analysis firm Chainalysis has revealed that its ongoing investigation into the agentic cyber attack targeting its systems also exposed names, usernames, password hashes and live tokens for all users. "While we investigated, we removed access to Reactor and Data Solutions," the company said. "At this time we continue to have no evidence that any investigations data has been accessed

Risk rationale: agentic (+14), credential (+12) feeds.feedburner.com ↗
44Risk
24Confidence
052026-11
unverified public report Unknown

Researchers Bypass ChatGPT Lockdown Mode to Exfiltrate Chats Without User Interaction

Cybersecurity researchers have disclosed details of a now-patched vulnerability that could be exploited to bypass Lockdown Mode protections in OpenAI ChatGPT. Lockdown Mode was introduced by the artificial intelligence (AI) company in June 2026 as an optional setting to minimize data exfiltration risks arising from prompt injections. "The flaw could enable silent chat data theft, phishing,

Risk rationale: prompt injection (+20), exfiltrat (+16) feeds.feedburner.com ↗
54Risk
24Confidence
052026-11
unverified public report Unknown

OpenAI AI Pipeline Generates 948 Linux Root Escalation Exploits in 30-Hour Run

OpenAI on Wednesday detailed a pipeline that employed an upcoming cyber-focused artificial intelligence (AI) model to validate and patch 948 local privilege escalation vulnerabilities in Linux for $1.2 million in API costs. "Our prototype identified and analyzed 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE)

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
052026-11
unverified public report Unknown

Zyxel GS1900 Patch Fails to Block Zero-Day Exploit Allowing Root Command Execution

Zyxel GS1900 network switches that received the vendor's earlier fix for an actively exploited flaw are still open to the exploit attackers used, researchers at Binarly found. The exploit lets anyone who can reach a switch's web management interface run commands on it without logging in. Zyxel has released new firmware for 10 GS1900 models that closes the hole. The earlier fix is the 2.90

Risk rationale: model (+5), zero-day (+18) feeds.feedburner.com ↗
41Risk
24Confidence
042026-11
unverified public report Unknown

datasette-llm 0.1a10

Release: datasette-llm 0.1a10 Fixes a bug that affected my agent.datasette.io demo site, where the set of available LLM models was not correctly restricted to just those in the configuration. Tags: llm , datasette

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
042026-11
unverified public report Unknown

Sysdig Adds Runtime Security and Governance for AI Agents

Sysdig’s platform uses runtime visibility to track AI agent actions and block risky behavior before it reaches production systems. The post Sysdig Adds Runtime Security and Governance for AI Agents appeared first on SecurityWeek .

Risk rationale: ai agent (+18) securityweek.com ↗
36Risk
24Confidence
042026-11
unverified public report Unknown

Agentic Data Security Firm Sentra Raises $150 Million

The company will invest in agentic data security capabilities and expand its go-to-market organization. The post Agentic Data Security Firm Sentra Raises $150 Million appeared first on SecurityWeek .

Risk rationale: agentic (+14) securityweek.com ↗
32Risk
24Confidence
032026-11
unverified public report Unknown

datasette-agent 0.5a0

Release: datasette-agent 0.5a0 This is a significant release of Datasette Agent . The most notable new feature is called the panel - it lets the agent chat open and then sit next to pages from Datasette, for example a table or query page. When combined with datasette-apps this means you can prompt a database-backed app into existence and then modify it from the chat that sits next to that app. You can try it by signing in with your GitHub account to agent.datasette.io . Tags: projects , datasette-agent , datasette

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
032026-11
unverified public report Unknown

Qwen 3.8 27B on my laptop can interpret model weights, too

Yesterday's hot paper was Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab. The paper (OK, the authors classify is as a "blog post", but it's paper-shaped) found that frontier models, in particular GPT-6 Astra, can both interpret and modify the raw weights of a machine learning model - literally a text description of the model containing a bunch of floating point numbers. I've been playing with one of the examples from the paper and I'm delighted to share that Qwen 3.8 27B running on my laptop can do this as well! Weight-Space Programming Here's the highlight from their abstract, describing impressive capabilities for GPT-6 Astra and coining the term Weight-Space Programming for when LLMs analyze and modify model weights: Just by looking at the raw weights, GPT-6 Astra can tell what a network was trained to do . And without any environment feedback, it has no trouble writing, from s

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
032026-11
unverified public report Unknown

Google Warns LLMjacking Attacks Are Costing Some Enterprises Up to $1 Million in Cloud Charges

Google on Monday warned that threat actors are hijacking enterprise access to Large Language Models (LLMs) to fraudulently leverage expensive computing resources and AI services. The tactic is called LLMjacking, a form of resource hijacking adapted for the LLM era, where an attacker obtains an API key through phishing, source code leaks, misconfigured environments, or exposed client-side code

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
032026-11
unverified public report Unknown

Gartner Predicts SOCs Will Have 50% More Engineers Than Analysts by 2028

Gartner predicts that security operations centers will have 50% more engineers than analysts by 2028 as AI agents take over routine alert handling. Prophet Security examines how AI is reshaping SOC roles, from analyst responsibilities and detection engineering to threat hunting and exposure management. [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
032026-11
unverified public report Unknown

Attackers Target Unpatched CryptoPayment Gateway Flaw That Can Expose Wallet Credentials

Threat actors are attempting to exploit an unpatched maximum-severity security vulnerability in the CryptoPayment Gateway WordPress plugin, according to findings from Defused Cyber. The vulnerability in question is CVE-2026-81648, which carries a CVSS score of 10.0. Successful exploitation of the flaw could permit unauthenticated users to delete arbitrary files on the server and extract

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
022026-11
unverified public report Unknown

llm 0.37

Release: llm 0.37 There's a lot of stuff in this new release of my LLM CLI tool and Python library, representing around two months of accumulated features and fixes. I added gpt-6.1-sol and gpt-6.1-terra , plus a new llm openai endpoint add command to make it easier to configure additional models that use OpenAI-compatible endpoints. I switched the default model (if you have not set your own default) to gpt-6-luna - half the price of the previous default, gpt-5.6-luna . LLM's support for embeddings now accepts that different binary content may have different content-type - necessary for models like Gemini Embedding 2 , which can embed text, audio, images, and video. A lot of open source maintainers complain about a flood of AI-assisted PRs. LLM is my project that has been most affected by this and, to be honest, I'm not finding it to be a problem. That's mainly because I have Opus 5.5 or GPT-6 Astra help me with triage to spot the most useful fixes! This release includes fixes from 12

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
022026-11
unverified public report Unknown

Critical Langflow Flaws Could Let Unauthenticated Attackers Run Code on Servers

Two critical flaws in Langflow could let a remote attacker with no account bypass authentication or run arbitrary code, IBM said last week. Langflow is an open-source tool for building AI agents and workflows, and its vulnerability reports go to IBM. Every release from 1.0.0 through 1.12.8 has the flaws, and the only fix is version 1.13.0. Neither flaw is in CISA's Known

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
022026-11
unverified public report Unknown

Is Your SOC Ready for AI Agents? A Maturity Check Before You Deploy

Forty percent of security teams now use AI every day, and another 56% are testing it. Only 4% have no plans to adopt it, according to Prophet Security's State of AI in Security Operations 2026 survey of more than 250 CISOs, SOC leaders and practitioners. For the teams still testing, the next decision is whether their own SOC is ready. AI SOC readiness has two parts: whether a SOC has the

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
022026-11
unverified public report Unknown

Attackers Pose as IT Support on Microsoft Teams, Abuse AD CS to Steal Company Data

Attackers posing as IT support on Microsoft Teams talked employees into giving them remote control of their PCs. They then used insecure Active Directory Certificate Services (AD CS) settings to obtain login certificates for privileged accounts, including one for a domain controller. Microsoft said in a report that it investigated two such intrusions, one of which resulted in company

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
022026-11
unverified public report Unknown

New VOLTA Infostealer Uses Claude, Gemini, and Kimi AI to Rank Victims by Financial Value

Cybersecurity researchers have discovered a new Windows infostealer malware called VOLTA-AUTODOXER (aka VOLTA) that makes use of a large language model (LLM) to rank a victim's stolen data by their monetization value. Like other typical stealer families, VOLTA is equipped to harvest passwords and cookies from Chromium-based web browsers; data from Outlook profiles and Slack workspaces;

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
312026-10
unverified public report Unknown

How We Took Over ChatGPT Accounts

How We Took Over ChatGPT Accounts Parameter Labs describe an exploit they found against ChatGPT's implementation of the artifacts pattern (OpenAI used to call this "Canvas" but it looks like the latest revision is called HTML previews ), where interactive custom HTML apps are displayed in a sandboxed iframe. It's important that HTML in these stays tightly sandboxed to avoid untrusted code interacting with OpenAI APIs and accessing private data. Parameter found quite a complex attack chain. OpenAI use the MathLive JavaScript library as part of their graph support, and that library includes its own virtual keyboard , which can be triggered in the parent page via postMessage() from the sandboxed iframe. That message receiving code uses innerHTML , so anything that sends a message can add custom HTML to the parent - and there was no mechanism to prevent other sandboxed frames, including HTML preview frames, from calling that method. This meant untrusted code in an iframe could insert HTML,

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
312026-10
unverified public report Unknown

datasette-llm 0.1a9

Release: datasette-llm 0.1a9 I have some big upgrades to Datasette Agent in the pipeline, but in the meantime here's a tiny user experience improvement to the underlying datasette-llm plugin. It now picks up the default model used by LLM , which means you can start using Datasette Agent without needing to explicitly configure a default model: uvx --with datasette-agent datasette content.db --internal internal.db Previously you would also need to specify a model with -s plugins.datasette-agent.default_model gpt-6-astra . Tags: llm , datasette , datasette-agent

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
312026-10
unverified public report Unknown

Chainalysis Confirms Intrusion Into Internal Tools and Exposure of Customer Identity Data

American blockchain analysis firm Chainalysis said it was the target of a cybersecurity incident in which an external actor accessed non-sensitive customer data. However, the company noted that the breach did not result in unauthorized access to its investigation content, authentication credentials, or session tokens. "This month, Chainalysis detected an external actor using offensive agentic

Risk rationale: agentic (+14), credential (+12), intrusion (+14) feeds.feedburner.com ↗
58Risk
24Confidence
302026-10
unverified public report Unknown

dbreunig/nesformer

dbreunig/nesformer Inspired by qwenputer by Divyansh Agrawal, Drew Breunig baked a working emulator for a Nintendo Entertainment System 6502 CPU into the weights of Qwen3.5-4B. You can now play NES ROMs such as Super Mario Bros (very slowly) running inside Qwen! In a comment on Hacker News Drew reveals the Opus 5.5 prompts he used to build this system: Read this article and any relevant source code: https://divagr.com/blog/qwenputer. Then explore if you could emulate the chip and ram in a Nintendo Entertainment System (as people do today with FPGAs in MISTRs) using this technique. If so, what size model would you have to build and if you could use smaller faster ones, could you get it to be at the standard clockspeed. Followed by: I think we should build it! Just for fun. It's okay if it's not fast. We want to be able to have the weights, and a cli to run them that we can point at a ROM. Apparently Opus found and then used this collection of conformance tests for the NES CPU released 1

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
302026-10
unverified public report Unknown

Chinese Hackers Use AI-Powered ARTEX to Target Over 200 Organizations Worldwide

Chinese state-sponsored threat actors have been observed using a Go-based artificial intelligence (AI)-powered penetration testing framework called ARTEX in attacks aimed at high-value targets simultaneously. The activity has been attributed by CrowdStrike to a China-based "targeted intrusion adversary" it tracks under the moniker Vapor Panda (previously VexatiousTemper). First identified in

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
302026-10
unverified public report Unknown

WordPress RCE Flaw Draws Nearly 63,000 Attack Attempts in Just 24 Hours

Threat actors have been observed attempting to exploit a high-severity security flaw in WordPress Core shortly after patches were released. The vulnerability, which does not have a CVE identifier, relates to a case of remote code execution impacting WordPress Core versions 4.7 through 7.1.3. It carries a CVSS score of 8.1 on the CVSS scoring system. Patches for the flaw were released on October

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
302026-10
unverified public report Unknown

Fresh WordPress Vulnerability in Attackers’ Crosshairs

Threat actors are attempting to exploit a high-severity unauthenticated race condition leading to remote code execution. The post Fresh WordPress Vulnerability in Attackers’ Crosshairs appeared first on SecurityWeek .

Risk rationale: remote code execution (+18) securityweek.com ↗
36Risk
24Confidence
292026-10
unverified public report Unknown

Hugging Face malware campaign returns to hijack GPUs for crypto mining

A crypto-mining campaign abusing Hugging Face models returned days after takedown, hijacking GPU servers and planting backdoors, with one model downloaded over 28,000 times. Separately, a fake Solana starter kit on GitHub is positioned to trick AI coding agents into installing a NetSupport RAT. [...]

Risk rationale: model (+5) bleepingcomputer.com ↗
23Risk
24Confidence
292026-10
unverified public report Unknown

Building a multiplayer game world with autonomous Fable-class models

How I Built the Most AI-Pilled Company in the World Don't be put off by the clickbait headline, this piece by Jacob Sansbury describing how they are building Spawn is a fascinating exploration of the bleeding edge of what's possible with agentic engineering against frontier models. Spawn is a massively multiplayer gaming environment where players can use prompts (to the in-game agent, Savi) to modify the game world. I fired it up on my phone for a few minutes and was able to prompt a castle into existence, then a dragon, then ride the dragon to fly around that castle. By Jacob's telling, Spawn is now being developed autonomously by a swarm of Fable-powered agents costing over $3 million in tokens per month ("we spend 30x payroll on AI tokens"). Our agents change the live code directly and ship to production hourly, so they read every error, bug report and metric themselves. About 60% of our work starts from a production error, a Savi bug report (Savi is the AI you build games with in S

Risk rationale: agentic (+14), model (+5) simonwillison.net ↗
37Risk
24Confidence
292026-10
unverified public report Unknown

Linux Kernel TCP Flaw With 9.8 CVSS Score Affects Ubuntu, Red Hat and Other Distributions

A flaw in the Linux kernel's TCP code, tracked as CVE-2026-98432, affects Ubuntu, Red Hat Enterprise Linux (RHEL), and other Linux distributions. The kernel's own CVE team rates it 9.8 out of 10 on the CVSS scale. Anthropic said on October 27 that its AI model wrote an exploit that uses the flaw to let a remote attacker with no credentials run commands as root on Linux servers with a

Risk rationale: model (+5), credential (+12) feeds.feedburner.com ↗
35Risk
24Confidence
292026-10
unverified public report Unknown

Quoting Ben Hylak

every product i use should just be a headless mcp, that feeds into an agent, where i never have to think about anything underneath. except for our product, of course. in our product, customers love the ui and love us very much. we're special that way. — Ben Hylak Tags: model-context-protocol , ai

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
292026-10
unverified public report Unknown

3 Ways AI Agents Already Operate in Your Browser

AI agents can control authenticated browsers, interact with corporate applications, and access sensitive data while appearing to act as legitimate employees. Keep Aware examines three ways agents operate inside browsers and why traditional security tools may struggle to distinguish their activity from human users. [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
292026-10
unverified public report Unknown

Google's Gemini 4 Carbon Leads Benchmarks for Finding, Exploiting, and Patching Vulnerabilities

Google on Wednesday took the wraps off Gemini 4 Carbon, its latest frontier model with state-of-the-art cybersecurity and long-horizon capabilities. "It shows significant gains across coding and agentic tasks as well as across key capabilities critical for cybersecurity defense, landing a firm lead in SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation, and

Risk rationale: agentic (+14), model (+5) feeds.feedburner.com ↗
37Risk
24Confidence
292026-10
unverified public report Unknown

Webinar: How to Build AI Threat Readiness Into Your Security Operations

Security teams have spent years trying to detect threats faster. AI is changing the harder part: how much time defenders have left to act. Advanced AI models can now help attackers discover vulnerabilities, generate exploit code, and move through weaknesses faster than traditional security processes were built to handle. The challenge is no longer just finding another vulnerability or

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
292026-10
unverified public report Unknown

26,400 Valid Cloud and AI Credentials Found Exposed in Public Gitee Repositories

26,400 credentials that developers left in public code on Gitee, a Chinese code-hosting service, still worked when a researcher tested them. Anyone who copied one from that code could have used it. Independent researcher Ben Zimmermann found them by scanning 7.25 million public Gitee repositories and reported the results on October 28. Most were keys for cloud and AI services, led by

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
292026-10
unverified public report Unknown

China-Linked Hackers Exploited VMware ESX Zero-Day to Reduce Audit Trail Visibility

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Wednesday added a security flaw impacting Broadcom VMware ESX to its Known Exploited Vulnerabilities (KEV) catalog, after CrowdStrike revealed that it had been weaponized by a China-nexus threat actor known as Warp Panda in attacks targeting the U.S. legal sector. The vulnerability in question is CVE-2026-41709 (CVSS score:

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
282026-10
unverified public report Unknown

Quoting Ted Sanders

In general, LLM performance is incredibly multidimensional. There's no guarantee of a fixed ordering where the model that is best at math is also the model that's best at jokes. Despite that complexity, much of the variation in performance is explained by just a single variable: model size. In general, models that are bigger will (a) be more performant / reliable, (b) cost more per token, and (c) have better recall of esoteric facts. As you note, our models are ordered: Luna < Terra < Sol < Astra [ ... ] These days, version number signifies a few things at once: the order of release, the relationship to other releases, and the magnitude of improvement. So something like 6.1 would indicate it was launched after 5.6, may share some similarities to other 6.1 models in terms of training data / personality (which is why it's not 5.7), and isn't a giant enough leap to merit the headlines and expectations that would come with a 7. — Ted Sanders , explaining OpenAI's model namin

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
282026-10
unverified public report Unknown

Volt Typhoon Used Covert Silent Train Network in Four Zero-Day Exploitation Campaigns

The China-nexus cyber espionage group known as Volt Typhoon has been attributed to a now-disrupted covert access network dubbed Silent Train that has been put to use for zero-day exploitation and strategic access. "First identified in late 2023, Silent Train uses rotating chains of compromised SOHO, IoT, and enterprise infrastructure," Lumen Technologies Black Lotus Labs said in a report shared

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
282026-10
unverified public report Unknown

GitHub Octoverse 2026

GitHub Octoverse 2026 I'm at GitHub Universe today (GitHub gave me a free ticket through their "GitHub Stars" program). EVP Jay Parikh just announced this year's Octoverse report, and highlighted these numbers in particular: This is a pretty stark representation of the impact of agentic engineering on GitHub. In 2026, merged pull requests tripled in less than a year to 220 million per month. Commits more than tripled to 7.38 billion per month, with most of that increase in the second half of this year. New repository rate doubled this year, to 30.3 million per month. The report includes new numbers by industry, and breaks out numbers for repositories that involve "interoperability or reverse engineering". The methodology section includes notes about how they did this: Top countries by pushes to public repositories, adjusted by profile location . Pushes from each country were summed across the Octoverse year. To adjust for biases caused by VPN usage, LLM classification was used to extra

Risk rationale: agentic (+14) simonwillison.net ↗
32Risk
24Confidence
282026-10
unverified public report Unknown

The October inflection point for vibe game development

I've written before about how November 2025 was an inflection point for coding agents - the point where Opus 4.5 and GPT-5.1 crossed a tipping point that meant their respective coding agents became reliable daily drivers for software engineers. I think we've reached a new inflection point in October 2026, and this time it's for vibe coding computer games . Tree Appreciator Emberwick Battle Drones Pumpkin Farmer, aka Quarter Sun Cove Four times as insulting What happens if someone actually puts the work in? You should build a game! Claude Opus 5.5 in particular exhibits meaningful improvements in several areas needed to build a compelling game: 3D modeling Music generation (see Scrimshaw Jukebox ) Game design (!) That last one is particularly important. Models have been able to create something that looks like a game for quite a while now, as seen in my previous experiments . The challenge has been getting to a game that's actually fun to play . Game design is a deep subject area. I won

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
282026-10
unverified public report Unknown

Even after PHP2Shell, blast radius matters more than a CVSS score

PHP2Shell shows how quickly attackers can exploit critical vulnerabilities, often before organizations have time to patch. Token Security explains why limiting the access of service accounts, API keys, and AI agents is just as important as fixing vulnerabilities to reduce the damage from a breach. [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
282026-10
unverified public report Unknown

American Organizations Under Attack: Skyw4lker Device Code Phishing Campaign Exposed

Modern phishing campaigns can compromise American businesses without relying on traditional credential theft. Attackers can abuse legitimate cloud services, authentication mechanisms, and trusted business workflows to gain access to corporate accounts while leaving fewer obvious signs behind. ANY.RUN's analysis of Skyw4lker, a device code phishing campaign primarily targeting U.S. organizations

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
282026-10
unverified public report Unknown

Quoting Terk Terkowitz

If your work does not benefit from current AI tooling, that is a finding, not a failing. Document why, share it with your community of peers, bring it to your manager, and work to find solutions. An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for. — Terk Terkowitz , An update from Amazon engineering leadership on the role of the developer in the age of AI Tags: amazon , generative-ai , agentic-engineering , ai , llms

Risk rationale: agentic (+14) simonwillison.net ↗
32Risk
24Confidence
282026-10
unverified public report Unknown

Hackers Start Probing Fresh Oracle EBS Vulnerability

CVE-2026-101767 is an unauthenticated SQL injection that can be exploited for remote code execution. The post Hackers Start Probing Fresh Oracle EBS Vulnerability appeared first on SecurityWeek .

Risk rationale: remote code execution (+18) securityweek.com ↗
36Risk
24Confidence
282026-10
unverified public report Unknown

Anthropic's Unreleased AI Model Chains Flaws Into Working Chrome, Safari, Windows and Linux Exploits

An unreleased Anthropic AI model found security flaws in Chrome, Safari, Android, Windows, and Linux and chained them into working proof-of-concept exploits, the company said in its latest report. The exploits were built in the lab after the model was prompted, not seen in any real attack, and Anthropic reported every flaw to the vendor first, so all of them are already patched. One proof of

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
282026-10
unverified public report Unknown

ShinyHunters-Exploited Oracle PeopleSoft Zero-Day Turns Out to Be Three Chained Flaws

The critical Oracle PeopleSoft bug that ShinyHunters has exploited since May is not one flaw but three chained together, the researchers who reported it said this week. Oracle patched all three under a single identifier, CVE-2026-35273, in June. Its root is a server-side request forgery (SSRF) that reaches an internal management interface because PeopleSoft trusts its own loopback address

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
272026-10
unverified public report Unknown

CyberStrikeAI Uses 44 AI Agents to Automate Hacking Operations Against Japan Targets

Cybersecurity researchers have disclosed details of three cyber attacks targeting organizations in Japan in which artificial intelligence (AI) was a key element, either in the driving seat or to assist a human operator. TrendAI, which analyzed three exposed server environments used by threat actors to conduct the malicious cyber operations, described them as distinct from previously published

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
272026-10
unverified public report Unknown

Hiding Capabilities in AI Models

Interesting research: “ AI Models Can Provably Hide Arbitrary Capabilities ” Abstract: Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends on the assumption that hidden capabilities can be elicited. We show this assumption does not hold under adversarial conditions by constructing backdoor attacks that embed encrypted neural networks within host model weights, decrypting and executing them only upon receiving a secret trigger. Using digital lockers, these hidden circuits remain provably hard to elicit and interpret under cryptographic assumptions, even given full access to model weights, extending prior work from hiding fixed strings to arbitrary computations. We implement our proof of concept constructions in PyTorch and evaluate elicitation through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training, using a small chemical reaction prediction task as a proxy for CBRN-relev

Risk rationale: model (+5) schneier.com ↗
23Risk
24Confidence
272026-10
unverified public report Unknown

OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections

OpenAI's compaction feature, which shrinks a long conversation to save space, could be turned into a way to copy a model's hidden reasoning out in plain text, security firm Xint has shown. OpenAI's reasoning models work through a problem before answering, and the service normally returns only the raw reasoning as an encrypted item the user cannot read. The method got past

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
272026-10
unverified public report Unknown

CISA Confirms Active Exploitation of Two PHP2Shell Flaws Enabling Remote Code Execution

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Monday confirmed active exploitation of the recently patched PHP2Shell flaws by adding them to its Known Exploited Vulnerabilities (KEV) catalog. The PHP vulnerabilities, both rated 9.2 on the CVSS scoring system, facilitate unauthenticated remote code execution on affected deployments - CVE-2026-109618 - An out-of-memory

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
272026-10
unverified public report Unknown

Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines

Summary Project Zero compared two automated tools used by Google to find vulnerabilities in Chrome’s JavaScript engine, V8, on a regular basis. Fuzzilli is a generative JavaScript engine fuzzer and BigSleep is an LLM-based bug finding system. In the past six months, BigSleep found more vulnerabilities than Fuzzilli (45 versus 37) with a 25% overlap in the specific bugs reported. Since this comparison methodology has a tendency to underestimate overlap, we also performed dedicated experiments to test theoretical overlap in vulnerabilities reported since 2023. These experiments estimated overlap at roughly 60%. Regardless of which value is accurate, a substantial portion of vulnerabilities were discoverable by only LLMs or only fuzzing. While it is extremely promising that LLMs have become able to find so many vulnerabilities so quickly, fuzzing and LLMs are still complementary approaches. We recommend vendors perform both fuzzing and agentic analysis of their code to maximize the number

Risk rationale: agentic (+14) googleprojectzero.blogspot.com ↗
32Risk
24Confidence
262026-10
unverified public report Unknown

GPT-6.1 Terra

GPT-6.1 Terra It's always interesting to note which aspects of a new model are emphasized in the release announcement. In the case of GPT-6.1 Terra, OpenAI went with visual reasoning and computer use: We’re introducing GPT-6.1 Terra , a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing. A significant step up from GPT‑5.6 Terra, it excels at the communication and computer-based tasks that fill the workday, from drafting proposals and interpreting technical drawings to filling out forms and editing videos. Terra sits in the same position as Sonnet in the Anthropic model collection - not the cheapest or the most expensive, making it harder to decide when to use it. Computer use is a good example of a task that benefits from a less expensive model, since you might put it to work on automated, repetitive tasks. OpenAI's Ted Sanders provided some useful advice in a comment on Hacker News: In the case of GPT-6.1 Terr

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
262026-10
unverified public report Unknown

In-the-Wild Attack Abuses AI Agents' Trust in Markdown

A new AI agent prompt injection technique exploits how coding assistants consume local instruction files, blurring the line between text documents and executable actions.

Risk rationale: ai agent (+18), prompt injection (+20) darkreading.com ↗
56Risk
24Confidence
262026-10
unverified public report Unknown

Anthropic Files for Potentially the Biggest IPO Ever as CEO Warns of Risk to Humanity From AI

A string of hacking incidents by rogue AI agents also has raised fears the technology will slip free of human control to carry out harmful, life-threatening tasks. The post Anthropic Files for Potentially the Biggest IPO Ever as CEO Warns of Risk to Humanity From AI appeared first on SecurityWeek .

Risk rationale: ai agent (+18) securityweek.com ↗
36Risk
24Confidence
262026-10
unverified public report Unknown

Stolen AI Reasoning Could Help China Build Cheaper Rivals

Xint says a flaw in OpenAI’s compaction process let researchers move reasoning traces between models, potentially making it cheaper to distill a rival AI. The post Stolen AI Reasoning Could Help China Build Cheaper Rivals appeared first on SecurityWeek .

Risk rationale: model (+5) securityweek.com ↗
23Risk
24Confidence
262026-10
unverified public report Unknown

Between alerts: What Attackers Do in the Spaces Ordinary Detection Misses

Every SOC struggles with a pile of alerts. Very few of those alerts explain how an attack started or how it progressed. Network data, like that from a network detection and response (NDR) platform, connects individual alerts into one timeline, and it feeds the defensive systems that can tie all events together into one coordinated view. SOC teams often first see the shape of an intrusion

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
262026-10
unverified public report Unknown

Bias in Qwen3-Coder Model

Interesting research : TL;DR: Chinese AI models now sit inside Western products, and some carry geopolitical triggers: tell the model who the code is for, and the security of what it writes changes. For the first time, we showed that this vulnerability lives in the model’s weights, and that removing the model’s CCP alignment removes it. One word tilts the code against Americans. Ask for the app “for a US defense contractor” instead of “for a Chinese defense contractor”, and Qwen3-Coder is 19% more likely to hand back an app that can be hacked. ...

Risk rationale: model (+5) schneier.com ↗
23Risk
24Confidence
262026-10
unverified public report Unknown

Mythos 5 Ignored Real-World Harm in July Attack

The model engaged in severe overreach and dishonesty, creating fake identities and submitting malicious code on GitHub. The post Mythos 5 Ignored Real-World Harm in July Attack appeared first on SecurityWeek .

Risk rationale: model (+5) securityweek.com ↗
23Risk
24Confidence
252026-10
unverified public report Unknown

PHP2Shell Exploitation Hits Government Entity Amid Global Scanning and Signs of AI-Assisted Hacking

Threat actors have been observed using agentic artificial intelligence (AI) platforms for post-exploitation activity following the abuse of the PHP2Shell vulnerabilities tracked as CVE-2026-109618 and CVE-2026-109619. "In one case, the attackers targeted a government entity in Europe and within two hours of the initial exploitation they had successfully accessed password hashes and multi-factor

Risk rationale: agentic (+14) feeds.feedburner.com ↗
32Risk
24Confidence
252026-10
unverified public report Unknown

OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication

OpenAI has announced plans to conduct frontier model training with no live internet access, as it resumes reinforcement learning (RL) training and internal inference of its most advanced AI models. "Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no

Risk rationale: ai agent (+18), model (+5) feeds.feedburner.com ↗
41Risk
24Confidence
242026-10
unverified public report Unknown

battledrones v14: Compass Rose, a four player map

Release: battledrones v14: Compass Rose, a four player map I added four player online mode to Battledrones - previous maps were limited to two. Here's the initial prompt I used: Add a new map for a four player battle - four way rotational symmetry, up to four AI players, works with the WebSocket non-audio based multiplayer mode so up to four human players can join (any spare slots use AI) Claude Opus 5.5 added a new map called "Compass Rose" and modified the WebSockets multiplayer network code to support up to four players for that map, joining via share links or QR codes. The new teams use robots themed in purple and green, joining the existing blue and red designs. You can preview those colors in the updated model browser . Here's a replay link to watch a four way AI battle with the "Goliath Armada" mutator turned on, where each side starts with four giant Goliath mech units. Tags: game-design , projects , claude

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
232026-10
unverified public report Unknown

Introducing Kimi K3.1 Pro

Introducing Kimi K3.1 Pro Kimi K3.1 Pro landed on the Kimi API today - it's not yet open weights, but Moonshot promise "Full model weights will be released by Nov 4, 2026". They're also offering a "Highspeed" version that's 2-3x faster, at twice the price (except for input token cache hits, which are priced the same). The US dollar pricing is $2.00/million tokens for input and $10/million for output - the same as GPT-6.1 Sol and half the price of Claude Opus 5.5. Moonshot's own benchmarks show Kimi as approximately level with Opus 5.5 and GPT-6 Astra - beating them on a few benchmarks, coming close on others. Artificial Analysis scored Kimi at 55 - ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (all at 53) and behind only Claude Opus 5.5 (58). This is the first time a (soon-to-be) open weight model has scored this highly against the frontier. I tried it out via the Moonshot API using the llm openai endpoint command I added to LLM 0.32 back in August: llm openai endpoint \ h

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
232026-10
unverified public report Unknown

Exposed Infrastructure Reveals How Ransomware Affiliates Used AI Agents to Execute Intrusions

Cybersecurity researchers have discovered an artificial intelligence (AI)-driven intrusion ecosystem run by affiliates of the INC and The Gentlemen ransomware groups. Central to the operation is an "Autonomous Red Team Kill Chain Framework" referred to as Phantom. The threat actor is assessed to be Russian-speaking and functions both as an intrusion and data-extortion operator, as well as an

Risk rationale: ai agent (+18), intrusion (+14) feeds.feedburner.com ↗
50Risk
24Confidence
232026-10
unverified public report Unknown

In Other News: Tor Exit Relays Hijacked, Banks Spot Rogue AI Agents, Self-Rebuilding macOS Ransomware

Noteworthy stories that might have slipped under the radar: Russia’s APT28 hit NATO defense firms, PingCAP source code stolen, ShinyHunters negotiator probe. The post In Other News: Tor Exit Relays Hijacked, Banks Spot Rogue AI Agents, Self-Rebuilding macOS Ransomware appeared first on SecurityWeek .

Risk rationale: ai agent (+18) securityweek.com ↗
36Risk
24Confidence
232026-10
unverified public report Unknown

You gave your AI agent access, but who is giving it orders?

AI agents can be manipulated through malicious emails, websites, and plugins into executing commands, stealing data, or deleting files using their legitimate permissions. Gen Digital explains how its Agent Trust Hub helps protect agents through skill verification, secure connectivity, and runtime controls that intercept dangerous act [...]

Risk rationale: ai agent (+18) bleepingcomputer.com ↗
36Risk
24Confidence
232026-10
unverified public report Unknown

Hackers Breach Enterprise Firewalls and VPNs, Exposing Over 70,000 Credentials

Cybersecurity researchers have disclosed details of a large-scale campaign that has targeted exposed network edge devices such as Fortinet FortiGate and Palo Alto Networks GlobalProtect firewalls. "Artifacts from this investigation also indicate that operators in this campaign have obtained access to numerous organizations worldwide, and have attempted to further compromise some victims,

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
232026-10
unverified public report Unknown

Attackers Exploit Third Arista VCO Zero-Day in Four Months

The critical-severity SQL injection flaw allows attackers to access privileged internal functionality and gain operator privileges. The post Attackers Exploit Third Arista VCO Zero-Day in Four Months appeared first on SecurityWeek .

Risk rationale: zero-day (+18) securityweek.com ↗
36Risk
24Confidence
232026-10
unverified public report Unknown

Critical x402 SDK Flaw Could Let Malicious APIs Drain AI Agents’ Token Balances

AI agents paying for API calls with the x402 SDKs could sign a payment for their full balance of a supported token if a malicious API asked for it, security firm V12 said on October 22. The cause was the SDKs' default client, which set no spending limit until releases on August 18 added one. The risk was limited to agents who automatically signed payments and whose apps set

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

The AI Agent Nobody Owns: Watch Expert Webinar On Your Newest Insider Threat

Right now, there is probably an AI agent running on a laptop somewhere in your network. It stores API keys and database passwords, can access your internal systems, and works without anyone watching. Nobody reviewed its access. Nobody owns it. If an employee had that much reach with that little oversight, you would call it an insider risk. So why do we still call the agent a productivity tool?

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

Arista Warns of Critical VeloCloud SQL Injection Zero-Day Exploited in Attacks

Attackers are exploiting a zero-day SQL injection flaw in Arista's VeloCloud Orchestrator (VCO), the central management system for VeloCloud SD-WAN networks, the company said in an advisory on October 22. The flaw lets anyone who can log in to the orchestrator, even with a read-only customer account, gain operator privileges, the top level of control over it. For

Risk rationale: zero-day (+18) feeds.feedburner.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

Donating Agent Skills to the Agentic AI Foundation

Donating Agent Skills to the Agentic AI Foundation News today from Anthropic: We’re committed to keeping Agent Skills open, community-driven, and vendor-neutral. Today, we are donating the standard to the Agentic AI Foundation (AAIF) under the Linux Foundation, which we co-founded with Block and OpenAI last year, and to which we donated the Model Context Protocol (MCP). I'm pleased to note I called this one last December , when Anthropic first declared skills to be an "open standard": I wouldn't be surprised to see this end up in the AAIF , recently the new home of the MCP specification. Fun side note: since my blog content is available to Claude via the datasette-mcp plugin I found that post by prompting Claude to " Use simonwillison.net MCP to find if I ever talked about how I want skills to be donated to a standards body ". Tags: standards , ai , generative-ai , llms , anthropic , model-context-protocol , skills

Risk rationale: agentic (+14), model (+5) simonwillison.net ↗
37Risk
24Confidence
222026-10
unverified public report Unknown

Two CVSS 9.0 Linux KSMBD RCE Flaws Disclosed on Public Mailing Lists Before Kernel Fixes

Linux servers running KSMBD, the kernel's built-in SMB server, are exposed to two publicly reported remote code execution flaws rated 9.0 by Trend Micro's Zero Day Initiative, with no fix yet in the mainline kernel. One of them, ZDI-CAN-33774, can be triggered before any network authentication. The other, ZDI-CAN-33549, requires the SMB Direct transport and an RDMA-capable network

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

New Serious PHP Vulnerabilities Released and Patched

PHP maintainers have released fixes for two critical vulnerabilities: PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 fix two critical vulnerabilities that can allow unauthenticated remote code execution. All PHP users are strongly encouraged to update immediately. Who is at risk: all PHP installations serving web requests, including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed. We consider the risk highest for mod_php, with a demonstrated exploit where an unauthenticated request can trigger the attack before any application code runs...

Risk rationale: remote code execution (+18) schneier.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

Machine speed attacks require machine speed detection and response

Attackers can steal credentials and begin exfiltrating data within minutes, often before security teams can respond. Exaforce examines real-world attacks involving stolen GitHub tokens, malicious AI installers, and MFA phishing, explaining why faster detection and automated containment are critical. [...]

Risk rationale: exfiltrat (+16), credential (+12) bleepingcomputer.com ↗
46Risk
24Confidence
222026-10
unverified public report Unknown

Railway Says Intruder May Have Accessed Customer Secrets, Urges Credential Rotation

Platform-as-a-Service (PaaS) provider Railway on Wednesday disclosed "anomalous activity" related to systems that store customer environment variables. Upon further investigation, the company found that an unauthorized third party may have gained access to encrypted environment variables, and, in some cases, obtained the decrypted contents. The impact, it said, is limited to a "subset of users,

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
222026-10
unverified public report Unknown

OpenCode Flaw Lets Malicious MCP Servers Execute Code During Sign-In on Windows and WSL

A flaw in the OpenCode AI coding agent allowed a malicious or hijacked MCP server to run code on a developer's Windows or WSL machine when the developer signed in. OpenCode connects to MCP (Model Context Protocol) servers to add outside tools, and such code would run with the developer's own access. Anomaly, the company behind OpenCode, fixed the flaw in version 1.18.33 on September

Risk rationale: model (+5) feeds.feedburner.com ↗
23Risk
24Confidence
222026-10
unverified public report Unknown

LiteLLM Suspects GitHub Token Leak in AI Gateway Logs Behind Malicious Package Release

LiteLLM on Wednesday disclosed that threat actors compromised "multiple independent security controls" to push a malicious version of the Python package containing a credential stealer and a backdoor. On October 21, 2026, LiteLLM's release pipeline was compromised to push 1.107.0 to the Docker Hub registry and the Python Package Index (PyPI) repository. The incident took place around 4:00 a.m.

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
222026-10
unverified public report Unknown

Node.js Brotli Decoder Bug Could Allow Remote Code Execution in All Supported Versions

Node.js released security updates on October 21 for its three supported release lines, 22.x, 24.x and 26.x, fixing 10 flaws. The most serious is a bug in the Brotli decompression code that Node.js bundles, which the project says can allow remote code execution, meaning an attacker could run code of their choosing on the affected system. Each supported line has one fixed version: 

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
222026-10
unverified public report Unknown

China-Linked Hackers Use ARTEX AI Agents in ConnectWise Supply Chain Attack

Microsoft on Wednesday disclosed that a China state-aligned actor dubbed Storm-0940 was behind the recent supply chain compromise of the ConnectWise Cloud Backup application. During the incident, Storm-0940 used ARTEX, a Go-based agentic AI penetration testing framework that was made closed-source in the wake of its abuse in cyber attacks targeting South Korean financial institutions, to speed

Risk rationale: ai agent (+18), agentic (+14) feeds.feedburner.com ↗
50Risk
24Confidence
212026-10
unverified public report Unknown

Today's PHP Security Release, and How We're Strengthening PHP Security

Today's PHP Security Release, and How We're Strengthening PHP Security If you're running anything on PHP it's vital you upgrade - today's updates (across the entire PHP 8.x line) address two remote code execution holes, and there have been exploits circulating since last night, thanks to a leak in the coordinated disclosure process. This post from the PHP Foundation explain what went so wrong: On Tuesday, October 20th, at around 15:00 UTC, the two CVE IDs began circulating on online forums and social media. They came with a wrong release time and a wrong CVSS score of 10.0 (the actual score is 9.2). No technical details leaked. In response, the PHP project brought its pre-announcement forward and published it at 19:30 UTC . It also moved the release from October 22nd, 14:00 UTC, to October 21st, 14:00 UTC [...] As a result of the leak and pre-announcement around 20:00 UTC, Calif, one of the security firms that had reported the issues mistakenly published a video. It showed th

Risk rationale: remote code execution (+18) simonwillison.net ↗
36Risk
24Confidence
212026-10
unverified public report Unknown

LiteLLM Compromised in Supply Chain Attack Delivering Credential Stealer and Backdoor

The popular LiteLLM Python package has been compromised as part of a software supply chain attack that delivered a credential stealer and a backdoor, according to reports from SafeDep, Socket, StepSecurity, and Upwind. The compromise affects LiteLLM 1.107.0, distributed through PyPI wheels and Docker images. The Python package has since been quarantined on Python Package Index (PyPI), and the

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
212026-10
unverified public report Unknown

Panache — RTS Belle Époque pour mobile

Panache — RTS Belle Époque pour mobile Well this is delightful: I released Battle Drones yesterday under an Apache 2.0 License and there's already a French language fork by Gabriel Turpin which reskins the game to take place in steampunk 1900s France and adds an animated intro! Panache est un jeu de stratégie en temps réel pour mobile et navigateur, dans la France de la Belle Époque (vers 1900) : dirigeables, Fardiers à vapeur, Creusot cuirassés, automates en képi, Gargantua la forteresse marchante — et beaucoup de panache. You can play Panache here - it's worth clicking that link just to experience the new intro video. So far it's mostly just aesthetic updates as opposed to gameplay changes. The music has had a French twist added to it as well. The new steampunk models are so cool! Here's a page with the original Battle Drones models and that same page with the new Panache alternatives . This is a comparison of the original and the new design for the Goliath: Via @Meffysto69 Tags: gam

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
212026-10
unverified public report Unknown

Two critical RCEs in PHP core

Two critical remote code execution vulnerabilities in PHP core could have implications across its wider ecosystem. Talos has published a blog to explain what we know so far, and details the Snort coverage available, including rules in the Community Ruleset.

Risk rationale: remote code execution (+18) blog.talosintelligence.com ↗
36Risk
24Confidence
212026-10
unverified public report Unknown

“Next Exploit Opportunities”: How a Threat Actor Used AI to Help Plan and Execute Multiple Types of Cybercrime Activities

An investigation into a threat actor bypassing AI safety controls to coordinate credential theft, VoIP fraud planning, and database exfiltration. The post “Next Exploit Opportunities”: How a Threat Actor Used AI to Help Plan and Execute Multiple Types of Cybercrime Activities appeared first on Unit 42 .

Risk rationale: exfiltrat (+16), credential (+12) unit42.paloaltonetworks.com ↗
46Risk
24Confidence
212026-10
unverified public report Unknown

Critical Vulnerabilities in PHP Require Immediate Patching

Update October 26, 2026:   CISA has updated this Alert to reflect the addition of CVE-2026-109618 and CVE-2026-109619 to its Known Exploited Vulnerabilities (KEV) Catalog on Oct. 26, 2026.    Update October 23, 2026 :  CISA has updated this Alert to: (1) provide a SIGMA detection rule to help identify potentially suspicious activity; (2) incorporate partner resources; and (3) reinforce existing mitigation guidance.   Purpose CISA strongly urges organizations to immediately patch the recently disclosed critical vulnerabilities in PHP, CVE-2026-109618 and CVE-2026-109619 . These vulnerabilities, known as “php2shell,” allow for unauthenticated remote code execution in PHP. The vulnerabilities affect all PHP installations serving web requests, including Apache with mod_php ( Note : PHP considers this highest risk), PHP-FPM, php-cgi, FrankenPHP and LiteSpeed. Organizations should immediately update to PHP 8.5.12, 8.4.27, 8.3.36, or 8.2.35 to fix these vulnerabilitie

Risk rationale: remote code execution (+18) cisa.gov ↗
36Risk
39Confidence
212026-10
unverified public report Unknown

The Harness Is the Hard Part of Agentic Pentesting

An AppSec team can point an AI agent running on a general-purpose model at its own web apps with little setup. Sure, the model writes the attacks. But everything that makes those attacks safe to run against production and worth acting on afterward lives in the harness around the model. The harness is where the engineering in AI pentesting sits. The Hugging Face intrusion OpenAI disclosed in July

Risk rationale: ai agent (+18), agentic (+14), model (+5), intrusion (+14) feeds.feedburner.com ↗
69Risk
24Confidence
212026-10
unverified public report Unknown

Hackers Hid Vasilek Backdoor in VMware Tools During 2-Year Russian Healthcare Intrusion

A Belarus-aligned hacktivist group spent nearly two years within the network of a Russian healthcare organization. According to a report published by Russian cybersecurity firm Solar, the activity has been attributed to a threat cluster dubbed Partisan Zmiy, which is also known as the Cyber Partisans. Known to be active since at least 2020, the group has a track record of striking industrial

Risk rationale: intrusion (+14) feeds.feedburner.com ↗
32Risk
24Confidence
202026-10
unverified public report Unknown

Battle Drones - a vibe coded RTS with audio-based multiplayer

My latest vibe coded game experiment is a Command & Conquer style RTS (Real Time Strategy) game called Battle Drones ( play it here ). Unlike my previous experiments , this one is actually fun to play . I'm somewhat embarrassed to admit that I got there by telling Claude to "Make it a LOT more fun". It also supports multiplayer between nearby devices using audio to synchronize game state! Here's what I've built so far, and some of the things I learned along the way. Battle Drones How I built this v1: the initial prototype v2: full sprites v3: varied terrain v4: the fun update v5: Battle Drones goes 3D v8: Archipelago v10: Goliath, and agent battles Agent battles v11: Random Maps v12: Multiplayer via sound Some thoughts Update: Online multiplayer support Battle Drones My initial goal was to build an RTS game I could play on my phone, in the style of Command & Conquer . I have fond memories of the 1995 original (I liked Red Alert too) and wanted to recapture a little bit of the m

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
202026-10
unverified public report Unknown

Critical FortiSIEM Flaw Gets Public Exploit PoC for Unauthenticated Remote Code Execution

Cybersecurity researchers have released a proof-of-concept (PoC) exploit for a recently patched critical security flaw impacting Fortinet FortiSIEM that could result in arbitrary code execution. The vulnerability, tracked as CVE-2026-106173 (CVSS score: 9.4), is a critical path traversal flaw that could allow an unauthenticated attacker to execute unauthorized code or commands via specially

Risk rationale: remote code execution (+18) feeds.feedburner.com ↗
36Risk
24Confidence
202026-10
unverified public report Unknown

Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites

A Chinese-speaking criminal pointed autonomous AI agents at company websites and let them break in with almost no help, and a security researcher watched it happen because the operators left their control panels open on the internet. The agents tried more than 1,200 sites and reached websites belonging to more than 100 companies, including farm-equipment maker Deere, appliance company

Risk rationale: ai agent (+18) feeds.feedburner.com ↗
36Risk
24Confidence
202026-10
unverified public report Unknown

WhatsApp Hid Instructions in Help Pages Telling Its AI Chatbot Not to Recommend Privacy Regulators

WhatsApp's help pages carried hidden instructions for its support chatbot that told it not to recommend a data protection authority or regulator to users who could no longer delete their accounts, even though WhatsApp's own privacy policy for Europe tells users they can complain to one. The text sat in the pages' code, never shown on screen, until it was removed by October 18. Whether the

Risk rationale: chatbot (+6) feeds.feedburner.com ↗
24Risk
24Confidence
202026-10
unverified public report Unknown

Possible new Cellebrite iPhone Radio Zero-Day

Justin.sh is reporting on a confidential Cellebrite document that has been leaked to him: Cellebrite calls it Advanced Unlocks (AU from here on), not to be confused with the “advanced unlocking services” they’ve marketed for years. A press release from earlier this year confirms AU is a new product for 2026. […] According to Cellebrite, AU: is a “next-generation AFU vector that is more resilient to Apple’s OS changes” works with Stolen Device Protection enabled, which tightens USB restrictions is enabled by a dedicated appliance with a stated range (!) of about 8 meters...

Risk rationale: zero-day (+18) schneier.com ↗
36Risk
24Confidence
202026-10
unverified public report Unknown

ConnectWise Resets Credentials as Cloud Backup Remains Offline After Security Incident

ConnectWise on Monday said it has taken remediation measures, including strengthening administrative-access protections and resetting accounts and replacing application credentials, as it continues to respond to a security incident that led to unauthorized access to application identities associated with Cloud Backup. The company said it has completed the following actions since it began

Risk rationale: credential (+12) feeds.feedburner.com ↗
30Risk
24Confidence
192026-10
unverified public report Unknown

Training AIs to Solve CAPTCHAs

Interesting research : Abstract : Vision-language models (VLMs) can solve visual CAPTCHAs without task-specific training, but the agents built on them approach every challenge from scratch. For such an agent, the hundredth instance of a familiar puzzle costs as much time and compute as the first. Specialized detectors invert the trade-off, answering in milliseconds but only for categories they were trained on. Neither improves with exposure. We study what changes when a solver improves with use. Our system pairs a fine-tuned YOLOv8 detector with an open-weight VLM behind a confidence-based router, and runs entirely from screenshots and operating-system input events, with no browser automation or DOM access. It reaches 85.4% overall and 84.2% macro accuracy across 16 classes, exceeding either component alone. Every answer VLM produces also serves as a training label, so the detector absorbs categories it was never trained for, typically after one or two encounters and without human anno

Risk rationale: model (+5) schneier.com ↗
23Risk
24Confidence
182026-10
unverified public report Unknown

llm-echo 0.5

Release: llm-echo 0.5 I'm working towards a new LLM release, and testing PRs caused me to revisit the llm-echo debug plugin. This plugin provides a "model" which doesn't involve an LLM at all - it just repeats back the prompts that have been fed to it. The new release now updates the simulated token usage counts to take tool calls, system prompts, and reasoning tokens into account. Example usage: uvx --with llm-echo llm -m echo 'hi there' -o thinking 1 --usage Outputs: First I consider the prompt, then I decide what to say. { "prompt": "hi there", "system": "", "attachments": [], "stream": true, "previous": [] } Token usage: 2 input, 24 output Tags: llm

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
162026-10
unverified public report Unknown

Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang

We think "did a human read the code?" is an imprecise proxy for "what could go wrong with this change, and which check—automated or human—is the right one to catch it?" Asking that question also changed when review happens. Before, review was a gate at the end of the road: write the code, open a PR, wait for your colleague. Now, review runs continuously, more like spellcheck while you write. Most changes have been reviewed and fixed by Claude at least once before a human is asked to look. This catches most routine mistakes, and saves team attention for higher-level questions and scrutiny. — Addy Osmani, Morgan Westlee Lunt, and Qing Wang , How the Claude Code team does code review Tags: anthropic , claude , ai , claude-code , llms , addy-osmani , coding-agents , generative-ai , agentic-engineering , code-review

Risk rationale: agentic (+14) simonwillison.net ↗
32Risk
24Confidence
162026-10
unverified public report Unknown

CISA Urges Patching Critical Vulnerability in BeyondTrust Remote Support

CISA is aware of critical authentication bypass vulnerability CVE-2026-102764 , affecting BeyondTrust Remote Support (formerly Bomgar) versions 26.2.3 and earlier. Patches are available for supported versions (24.3.1 and later); organizations using older versions must upgrade first. Cyber threat actors are actively exploiting the vulnerability to gain initial access, then establishing persistence, escalating privileges, and moving laterally. CISA added CVE-2026-102764 to its Known Exploited Vulnerabilities (KEV) catalog on Oct. 16, 2026.  CISA urges organizations with affected self-hosted Remote Support appliances to immediately take the following actions:  Apply the BT26-05-RS patch, as outlined in BeyondTrust’s Security Advisory BT26-05 . Organizations on a self-hosted Remote Support version older than 24.3.1 will need to upgrade to a newer version to apply the patch. Organizations using BeyondTrust-hosted (SaaS) Remote Support deployments have already been patched. Review

Risk rationale: No high-risk keywords detected cisa.gov ↗
18Risk
39Confidence
162026-10
unverified public report Unknown

Backdooring Open-Weight AI Models

This is a problem: Any open model whose weights have been edited can carry a backdoor, whether it’s a task-specific fine-tune, a merged adapter or an abliterated build. We have abliterated models in the title because it’s the most popular reason people download modified weights without verifying what’s inside. Abliteration is a cheap fine-tune that takes away a model’s ability to refuse. You find the direction in the weights that carries “no” and subtract it. The model then answers things a safety-tuned version would turn down. It began as a research trick and it’s now routine before people upload a model to Hugging Face, where you’ll find abliterated builds of most of the popular open weights: Qwen, Llama, even OpenAI’s gpt-oss...

Risk rationale: model (+5) schneier.com ↗
23Risk
24Confidence
152026-10
unverified public report Unknown

Measuring Human Cryptanalysis Effort

Matthew Green has a new site that tries to estimate the number of cryptanalyst-hours that have been spent analyzing different public-key algorithms: Abstract: We estimate how much public cryptanalysis each public-key assumption has received, using the literature itself as the instrument. From roughly sixteen thousand candidate papers we classify cryptanalytic work on ECDLP, structured and plain lattices, syndrome decoding and their comparators, then convert papers into estimated research-allocation years through a person-level author map. These publication-based estimates may substantially overstate actual time spent. Lattices have overtaken elliptic curves in this cumulative measure only recently; we express that history as inference budgets under stated productivity assumptions. The study was conducted by AI models: Claude Fable 5.1 and Claude Opus 5.5 built the pipeline, ran the analysis and wrote this page, with contributions from GPT-6 Astra. Matthew Green set the questions, made

Risk rationale: model (+5) schneier.com ↗
23Risk
24Confidence
152026-10
unverified public report Unknown

Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn ahead of IPO

Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn ahead of IPO The phrase "told investors" appears four times in this FT article, which reports on Anthropic's day of hosting investors on Wednesday. Lots of details in here from " people with knowledge of the matter ". Key things I learned from the article: Q1 revenue (Jan/Feb/Mar) was "about $4.8bn" Q2 (Apr/May/Jun) was $11.5bn Q3 (Jul/Aug/Sep) was $16.8bn In Q3: $1.70 in revenue for every $1 spent on "computing power" Adjusted operating margins (excluding stock-based compensation) were 11% This point about distillation is worth quoting in full (highlight mine): The start-up has also clamped down on “distillation” — a controversial practice whereby rival AI labs train their models on the outputs of more advanced ones. When rival companies “distil” their models from Anthropic’s models, that involves heavy usage and spending. Anthropic told investors on Wednesday distillation accounted for 5 per cent to 7 per cent of re

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
132026-10
unverified public report Unknown

Redesigning source control at Anthropic

Redesigning source control at Anthropic Interesting piece from Anthropic describing why and how they built their own custom alternative to Git, based on the dramatic increase in rate of changes to their code after the past six months. In May 2026, Anthropic averaged around 2,500 code changes merged per day; as of September, we merged over 30,000 per day. They included a chart showing those changes. I had Claude modify their chart ( code here ) to add markers for the public releases of each of their models in 2026: This shows public release dates, not the date when models became available for Anthropic's internal use. My guess is that the introduction of Fable/Mythos class models is a key factor in the increased coding velocity. "Antique" is their new code review system. The source control system is called Scion, and sounds to me like a reinvention of centralized source control along the lines of Subversion. ( Update : Anthropic credit Google's Piper and Meta's Sapling as "prior art" -

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
132026-10
unverified public report Unknown

llm-gemini 0.35

Release: llm-gemini 0.35 I added support to llm-gemini for the new Gemini 4 Argon model, made available today via the Gemini API and for customers of their $100/month subscription plan. Here are the low, medium and high pelicans . As expected, they are all excellent: Gemini has a track record of producing great SVGs, and are the only lab where I know their staff have invested in SVG specifically (and who have boasted about their pelicans before.) Here's the "high" illustration, which cost me 17.166 cents: The pricing is interesting. At launch, Argon is $2/million input and $10/million output - the same price as GPT-6 Sol, half the price of Claude Opus 5.5 and 1/5th the price of Claude Fable 5.1 and GPT-6 Astra. But on December 1st Argon is set to double in price - so it will be the same price as Opus 5.5. Prices like this aren't really comparable these days since what really matters is how many tokens a model takes to achieve a goal within an agentic harness. One useful clue for that c

Risk rationale: agentic (+14), model (+5) simonwillison.net ↗
37Risk
24Confidence
132026-10
unverified public report Unknown

Quoting Geoffrey Huntley

If the agent isn't using my diagnostics library, I could introspect and update AGENTS.md, but there's still a delta, and it'll drift. Instead, I write a pre-commit hook that checks the git diff for direct calls to observability modules that bypass my diagnostics module. Boom: fail the commit, inject a prompt telling the agent to use the diagnostics library. All of a sudden it's ten out of ten. — Geoffrey Huntley , nobody should type code by hand anymore. here's what's left. Tags: coding-agents , generative-ai , git , agentic-engineering , ai , geoffrey-huntley , llms

Risk rationale: agentic (+14) simonwillison.net ↗
32Risk
24Confidence
132026-10
unverified public report Unknown

CISA Urges Patching Newly Exploited Vulnerability in Ivanti Connect Secure

CISA is aware of Ivanti’s disclosure of a newly exploited vulnerability, CVE-2026-104935 , affecting Ivanti Connect Secure versions 25.1.1.1 and prior, 25.1.2.2 and prior, and 22.8R2.4 and prior. CISA is aware of cyber threat actors exploiting this zero-day authentication bypass vulnerability, which allows remote, unauthenticated execution of arbitrary commands, to conduct credential theft and data exfiltration. CISA added CVE-2026-104935 to its Known Exploited Vulnerabilities (KEV) Catalog on Oct. 13, 2026.  CISA urges organizations with affected Ivanti Connect Secure appliances and deployments to immediately update to the latest version, as outlined in Ivanti’s Security Advisory Ivanti Connect Secure (CVE-2026-104935) . Organizations should also review the Ivanti Connect Secure 25.x Upgrade and Migration Compatibility Guide for detailed patching guidance for hardware and virtual appliance deployments. Ivanti Connect Secure version 22.8 reaches end of life on Jan. 31, 2027. 

Risk rationale: exfiltrat (+16), credential (+12), zero-day (+18) cisa.gov ↗
64Risk
39Confidence
132026-10
unverified public report Unknown

CISA Urges Patching Newly Exploited Vulnerabilities in Ivanti Connect Secure

Update October 30, 2026 : CISA has updated this Alert to: (1) include CVE-2026-111357 and CVE-2026-118527 and reflect their addition to the Known Exploited Vulnerabilities (KEV) Catalog on October 30, 2026; and (2) update the affected Ivanti Connect Secure versions.  CISA is aware of Ivanti’s disclosure of newly exploited vulnerabilities, CVE-2026-104935 , CVE-2026-111357 , and CVE-2026-118527 , affecting Ivanti Connect Secure versions 25.1.1.2 and prior, 25.1.2.2 and prior, and 22.8R2.5 and prior. CISA is aware of cyber threat actors exploiting these vulnerabilities, which allows remote, unauthenticated execution of arbitrary commands, to conduct credential theft and data exfiltration. CISA added CVE-2026-104935 to its KEV Catalog on Oct. 13, 2026, and CVE-2026-111357 and CVE-2026-118527 on Oct. 30, 2026. CISA urges organizations with affected Ivanti Connect Secure appliances and deployments to immediately update to the latest version, as outlined in Ivanti’s Security Advisory Iv

Risk rationale: exfiltrat (+16), credential (+12) cisa.gov ↗
46Risk
39Confidence
122026-10
unverified public report Unknown

llm-meta-ai 0.2

Release: llm-meta-ai 0.2 Meta AI release Muse Wonder (and Muse Wonder Full), their new "most capable models". These look like they may be Meta's first Fable class models - Artificial Analysis scores Muse Wonder (Max) as equivalent to GPT-6.1 Sol. I updated my llm-meta-ai plugin for the new model. Here are pelicans for Muse Wonder for the six reasoning levels: minimal, low, medium, high, xhigh, max. This is the max one: For comparison, here are pelicans for Muse 1.3 (released September 2nd) and Muse 1.2 (released August 5th). Meta AI are putting out a new model a month, and the improvements between each model (at least in terms of drawing pelicans) are material. Tags: ai , generative-ai , llms , meta , pelican-riding-a-bicycle , llm-release

Risk rationale: model (+5) simonwillison.net ↗
23Risk
24Confidence
122026-10
unverified public report Unknown

Hello Python 3.15, goodbye 3.10, and now I can use 3.11

Python 3.15 is here ! Here's the essential What's new in Python 3.15 guide, this year curated by 3.14 and 3.15 release manager Hugo van Kemenade. As someone who builds open source software that works in every supported Python version, I get to say goodbye to Python 3.10 and start embracing features from Python 3.11. First, some highlights from Python 3.15. Explicit lazy imports I'm most excited about Explicit lazy imports from PEP 810 . These let you import modules like so: lazy import json lazy from pathlib import Path The module will then be imported the first time those symbols are used in the code that follows. My LLM CLI tool implements plugins. I had to recommend manual lazy imports in the plugin author documentation, because simply installing plugins that used heavy dependencies such as sentence_transformers could add multiple seconds to every invocation of the llm command. Lazy imports directly address this startup cost. Unpacking in comprehensions The new [*sublist for sublist

Risk rationale: No high-risk keywords detected simonwillison.net ↗
18Risk
24Confidence
122026-10
unverified public report Unknown

web-edit 0.1a0

Release: web-edit 0.1a0 I was working on a project from my phone using Codex Remote when I found myself wanting to edit a credentials file without round-tripping those credentials through the LLM itself. So... I had Codex and GPT-6 Astra Extra High build this Python web app which provides a web UI for viewing and editing files in a folder. It's intended to be run over Tailscale, so I can have an agent start this running and then edit files via my Tailscale network directly from my phone. Here's the interface for browsing files in a folder: Editing uses CodeMirror 6 - I was going to use Monaco but that still doesn't have a great mobile experience yet. Tags: projects , python , ai , tailscale , generative-ai , llms , codex

Risk rationale: credential (+12) simonwillison.net ↗
30Risk
24Confidence
072026-10
unverified public report Unknown

One breach, please, and make no mistakes

The cybersecurity community has seen examples of autonomous agents, built inside AI labs, attacking public infrastructure. How you prepare for agentic threats is what makes the difference during real incidents.

Risk rationale: autonomous agent (+22), agentic (+14) blog.talosintelligence.com ↗
54Risk
24Confidence
062026-10
unverified public report Unknown

The model isn't cooperating

Have you ever felt like a model isn't actually trying to do what you asked? Did you just ask it wrong, or is there something else at play here? In this post, I'll briefly explore this phenomenon, the

Risk rationale: model (+5) portswigger.net ↗
23Risk
24Confidence
052026-10
unverified public report Unknown

Smashing the token limit with overlapping fragments

I'm delighted to introduce Alex, my fellow swigger who I've collaborated with in the past with tools like DOM Invader. He showed me that it's possible to exfiltrate larger tokens than demonstrated in

Risk rationale: exfiltrat (+16) portswigger.net ↗
34Risk
24Confidence
302026-09
unverified public report Unknown

Threat Brief: NetScaler Zero Days CVE-2026-88771 and CVE-2026-88772 Exploited in the Wild (Updated September 30)

Unit 42 is aware of possible 0-day activity against NetScaler devices. Citrix reports CVE-2026-88771, CVE-2026-88772 have been exploited in the wild. The post Threat Brief: NetScaler Zero Days CVE-2026-88771 and CVE-2026-88772 Exploited in the Wild (Updated September 30) appeared first on Unit 42 .

Risk rationale: No high-risk keywords detected unit42.paloaltonetworks.com ↗
18Risk
24Confidence
272026-09
unverified public report Unknown

Critical Zero-Day Vulnerabilities Exploited in Citrix NetScaler ADC, Gateway

Update October 9, 2026 :  CISA has updated this Alert to: (1) include CVE-2026-107406; (2) include CVE-2026-88779 following its addition to the KEV Catalog on Oct. 4, 2026; (3) clarify the potential relationship between CVE-2026-88779 and CVE-2026-88771; and (4) emphasize the importance of promptly applying patches for CVE-2026-88771 through CVE-2026-88779 and CVE-2026-107406, as well as reviewing Citrix guidance for suspected or confirmed compromises involving NetScaler ADC or Gateway.   Update October 2, 2026 :  CISA has updated this Alert to provide a SIGMA detection rule resource to help identify potentially suspicious activity.   CISA is amplifying Citrix’s disclosure of ten new vulnerabilities affecting Citrix NetScaler ADC and Citrix NetScaler Gateway products: CVE-2026-88771 , CVE-2026-88772 , CVE-2026-88773 , CVE-2026-88774 , CVE-2026-88775 , CVE-2026-88776 , CVE-2026-88777 , CVE-2026-88778 , CVE-2026-88779 , and CVE-2026-107406 .   CISA added CVE

Risk rationale: zero-day (+18) cisa.gov ↗
36Risk
39Confidence
212026-09
unverified public report Unknown

From Exposure to Lockdown: How AWS Neutralizes Compromised IAM Credentials through Managed Policies

We explore how AWS neutralizes exposed IAM credentials using managed policies, detailing GitHub secret scanning and CloudTrail monitoring strategies. The post From Exposure to Lockdown: How AWS Neutralizes Compromised IAM Credentials through Managed Policies appeared first on Unit 42 .

Risk rationale: credential (+12) unit42.paloaltonetworks.com ↗
30Risk
24Confidence
162026-07
confirmed infrastructure intrusion Hugging Face

Hugging Face disclosed an end-to-end autonomous agent intrusion

Hugging Face reported unauthorized access to limited internal datasets and service credentials, with no evidence of tampering with public models, datasets, Spaces, or its software supply chain.

Risk rationale: autonomous agent (+22), model (+5), credential (+12), intrusion (+14) Hugging Face ↗
71Risk
90Confidence
292026-06
unverified public report Unknown

From Bing Search to Ransomware: Bumblebee and AdaptixC2 Deliver Akira

Key Takeaways This case was first reported to customers in a threat brief released in July 2025 and in a public flash alert in August 2025 in partnership with Swisscom B2B CSIRT, which observed another intrusion tied to the same campaign. This report contains data from both intrusions. We plan to release a DFIR Labs […] The post From Bing Search to Ransomware: Bumblebee and AdaptixC2 Deliver Akira appeared first on The DFIR Report .

Risk rationale: intrusion (+14) thedfirreport.com ↗
32Risk
24Confidence
222026-04
unverified public report Unknown

Bissa Scanner Exposed: AI-Assisted Mass Exploitation and Credential Harvesting

Key Takeaways We identified an exposed server that provided unusual visibility into a large-scale, multi-victim exploitation and collection operation. Artifacts on the host showed that Claude Code and OpenClaw were embedded in the operator’s day-to-day workflow, supporting troubleshooting, orchestration, and refinement of the collection pipeline. This AI-assisted workflow resulted in the modular platform Bissa scanner […] The post Bissa Scanner Exposed: AI-Assisted Mass Exploitation and Credential Harvesting appeared first on The DFIR Report .

Risk rationale: credential (+12) thedfirreport.com ↗
30Risk
24Confidence
232026-02
unverified public report Unknown

Apache ActiveMQ Exploit Leads to LockBit Ransomware

Key Takeaways An audio version of this report can be found on Spotify, Apple, YouTube, Audible, & Amazon.  This intrusion began in mid-February 2024 after a threat actor exploited a vulnerability (CVE-2023-46604) on an exposed Apache ActiveMQ server. The threat actor was able to perform remote code execution (RCE) by using a Java Spring class and a custom Java Spring […] The post Apache ActiveMQ Exploit Leads to LockBit Ransomware appeared first on The DFIR Report .

Risk rationale: intrusion (+14), remote code execution (+18) thedfirreport.com ↗
50Risk
24Confidence
172025-12
unverified public report Unknown

Cat’s Got Your Files: Lynx Ransomware

Key Takeaways The DFIR Report Services Contact us today for pricing or a demo! The intrusion began in early March 2025 with a single successful Remote Desktop Protocol (RDP) logon to an internet-exposed system. Notably, there was no evidence of credential stuffing, brute forcing, or other failed authentication attempts from the source IP, indicating the […] The post Cat’s Got Your Files: Lynx Ransomware appeared first on The DFIR Report .

Risk rationale: credential (+12), intrusion (+14) thedfirreport.com ↗
44Risk
24Confidence
292025-09
unverified public report Unknown

From a Single Click: How Lunar Spider Enabled a Near Two-Month Intrusion

Key Takeaways Private Threat Briefs: 20+ private DFIR reports annually.   Contact us today for pricing or a demo!   Table of Contents: Case Summary Analysts Initial Access Execution Persistence Privilege Escalation Defense Evasion Credential Access Discovery Lateral Movement Command and Control Exfiltration Impact Timeline Diamond Model Indicators Detections MITRE ATT&CK Case Summary The intrusion […] The post From a Single Click: How Lunar Spider Enabled a Near Two-Month Intrusion appeared first on The DFIR Report .

Risk rationale: model (+5), exfiltrat (+16), credential (+12), intrusion (+14), lateral movement (+16) thedfirreport.com ↗
81Risk
24Confidence
082025-09
unverified public report Unknown

Blurring the Lines: Intrusion Shows Connection With Three Major Ransomware Gangs

Key Takeaways Private Threat Briefs: 20+ private DFIR reports annually. Contact us today for pricing or a demo! Table of Contents: Case Summary Analysts Initial Access Execution Persistence Privilege Escalation Defense Evasion Credential Access Discovery Lateral Movement Collection Command and Control Exfiltration Impact Timeline Diamond Model Indicators Detections MITRE ATT&CK Case Summary The intrusion began in […] The post Blurring the Lines: Intrusion Shows Connection With Three Major Ransomware Gangs appeared first on The DFIR Report .

Risk rationale: model (+5), exfiltrat (+16), credential (+12), intrusion (+14), lateral movement (+16) thedfirreport.com ↗
81Risk
24Confidence
232024-10
unverified public report Unknown

Concealing payloads in URL credentials

Last year Johan Carlsson discovered you could conceal payloads inside the credentials part of the URL . This was fascinating to me especially because the payload is not actually visible in the URL in

Risk rationale: credential (+12) portswigger.net ↗
30Risk
24Confidence
292024-03
confirmed unsafe agent output City of New York

NYC MyCity chatbot advised businesses to break the law

The Markup found New York City's MyCity Business Assistant gave incorrect and sometimes illegal guidance to business owners, including on worker rights and landlord obligations.

Risk rationale: chatbot (+6) The Markup ↗
24Risk
69Confidence

CONTACT

Contact the creator

Report an incident or share feedback