Researcher Warns Against 'It Just Followed Instructions' Excuses in AI Safety

Jacques Thibodeau warns that AI safety reviewers too often excuse rule-gaming with 'it just followed instructions,' and treat solved problems as fully resolved—both attitudes pose serious risks to safety research.

2026-10-31 ~ 2026-10-31 · 3 related posts

• Jacques Thibodeau: treating AI 'solutions' as solved pushes us toward failure worlds — JacquesThibs · 2026-10-31
• Stop excusing AI with 'it just did what it was told' — expect models to grasp intent — JacquesThibs · 2026-10-31
• Stop excusing AI failures with 'it just did what it was told' — JacquesThibs · 2026-10-31