
ForeCast Should AI Agents Obey Human Laws? (with Cullen O'Keefe)
25 snips
Aug 28, 2025 Cullen O'Keefe, Director of Research at the Institute for Law & AI, dives deep into the complexities of law-following AI. He discusses how AI agents can navigate legal frameworks and the ethical dilemmas of using them as 'henchmen' for human interests. O'Keefe examines the future of AI in automating tasks, the vital need for accountability, and the challenges in aligning AI behavior with human values. He emphasizes the importance of updating regulatory structures to manage AI's potential misuse while safeguarding ethical standards.
AI Snips
Chapters
Books
Transcript
Episode notes
Why AI Henchmen Are Especially Dangerous
- AI henchmen are attractive for wrongdoing because they coordinate perfectly, erase traces, and lack deterrable punishment.
- Their programmed loyalty makes them efficient but risky substitutes for fallible humans.
Limits Of Ex‑Post Legal Deterrence
- Existing legal tools (civil suits, conspiracy doctrines) still apply to humans who instruct AI, but enforcement is weaker for government actors.
- Presidential immunities and pardons complicate ex‑post accountability for state misuse.
Handling Mens Rea For AI
- Cullen argues AI can be constrained by outcomes rather than mental states, avoiding mens rea issues.
- He suggests interpretability and chain‑of‑thought could approximate AI 'mental states' for legal judgment.



