Theo Jaffee and Ketan Ramakrishnan compare banking-style supervision with a release-gate model of AI regulation. Ketan Ramakrishnan argues that risks can emerge during training and internal deployment, so oversight should generate evidence throughout development rather than focus only on public release. He also warns that banking's confidentiality cannot simply be transplanted into politically consequential AI decisions.
Theo Jaffee and Ketan Ramakrishnan consider concentrating regulatory burdens on well-resourced frontier developers and designing cooperative testing and mitigation around limited public expertise. Ketan Ramakrishnan views spending-based institutional triggers as potentially more durable than technical model thresholds, while preserving democratic accountability for major risk trade-offs.
Theo Jaffee and Ketan Ramakrishnan then debate the strengths and limits of civil liability. Their discussion distinguishes reasonable-use harms from allegedly negligent or egregious model behavior and treats causation as a contested question, not a promise that a particular claim succeeds. Ketan Ramakrishnan argues that liability alone may discourage risk discovery and cannot replace institutions that generate evidence or require adaptive mitigation. These are the speakers' policy arguments, not current legal advice.
Watch on YouTube
