If you design a test to find misalignment, you will find misalignment
Risk rationale: ai agent (+18)
Source: theregister.com ↗
36Risk
24Confidence