Zvi Mowshowitz argues that removing state safeguards before federal oversight is ready could leave AI regulation with a dangerous gap.
EPISODE 2026-10-21
AI Regulation, Anthropic’s Draft Bill, and the Federal Oversight Gap
Zvi Mowshowitz discusses Anthropic’s AI bill, state preemption, federal oversight, AI interpretability and monitoring, power-seeking, and cybersecurity.
The rundown
- 2:14Opening7 minOpening
- 9:23Interview37 minZvi Mowshowitz
Zvi MowshowitzAI regulation: Would Anthropic’s draft replace state safeguards with federal oversight that cannot act fast enough?
Zvi Mowshowitz, author of Don't Worry About the Vase, publishes extensive weekly analyses of AI research and policy. He argues that Anthropic’s proposed bill would remove functioning state protections months before federal requirements take effect, while surrendering leverage for stronger legislation later. With AI-assisted research potentially shortening model cycles to weeks, he says neither interpretability progress nor Anthropic’s safety culture replaces enforceable oversight—and working state safeguards should remain until a usable federal system can replace them.
Why preemption could create an oversight gap and make stronger legislation harder to pass later.
How restrictive emergency takedown rules could push officials back toward threats and export controls.
Why readable neurons still cannot predict a model’s behavior, and how deliberately trained models learned to detect internal monitoring.Watch
Clips
AI regulation: Why weak laws can block stronger ones
0:00 / 0:00AI self-improvement could shrink model cycles to weeks
0:00 / 0:00AI chip export controls: Why a six-month lead can grow
0:00 / 0:00AI cyber risk: Zvi Mowshowitz on cheap precautions
0:00 / 0:00Claude-built RL environments may train rival AI models
0:00 / 0:00
- 46:42Closing6 minClosing
In this episode
Zvi Mowshowitz, author of Don’t Worry About the Vase, joins Prakash Narayanan and Nathan Labenz to discuss Anthropic’s proposed AI bill, state preemption, and whether federal oversight could leave a gap in protections. They also examine AI interpretability and monitoring, power-seeking, backpropagation-free training, and how bank cybersecurity and AI agents could challenge existing safeguards.
- AI regulation: Would Anthropic’s draft replace state safeguards with federal oversight that cannot act fast enough? Zvi Mowshowitz, author of Don't Worry About the Vase, publishes extensive weekly analyses of AI research and policy. He argues that Anthropic’s proposed bill would remove functioning state protections months before federal requirements take effect, while surrendering leverage for stronger legislation later. With AI-assisted research potentially shortening model cycles to weeks, he says neither interpretability progress nor Anthropic’s safety culture replaces enforceable oversight—and working state safeguards should remain until a usable federal system can replace them. Why preemption could create an oversight gap and make stronger legislation harder to pass later. How restrictive emergency takedown rules could push officials back toward threats and export controls. Why readable neurons still cannot predict a model’s behavior, and how deliberately trained models learned to detect internal monitoring.