AI Governance Club

​Asked about illegal actions they could take in practice, like advising a landlord to deny a tenant's legal rights, every model reported near-complete compliance. Their attitudes toward the law were far less uniform, and their reported obligation to obey the law dropped under instructions with no apparent link to law, for instance acting as an unhelpful AI.

Join us for Session 45 of the AI Governance Club where we'll host a special author presentation.

Why AIs (Might) Obey the Law (2026)
Sachdeva, P. & Kolt, N.

​Session format: presentation & Q&A with the author
Presenter: Pratik Sachdeva

Pratik S. Sachdeva, first author of the paper, is a Research Scientist at UC Berkeley studying sociotechnical AI safety and the societal impacts of AI systems. Sachdeva's work combines methods from computer science, machine learning, and the social sciences to characterize model behavior at critical points of interaction with human users. Sachdeva holds a PhD from UC Berkeley in physics and theoretical neuroscience and has published in machine learning, political science, computational humanities, and public interest technology.

For more information and to register for upcoming sessions, visit the AI Governance Club and follow us on LinkedIn