Harvard researcher dissects Sanders' ASI ban bill: too vague, ignores hardware supply chain
StephenLCasper · x · 2026-09-26
Harvard researcher Stephen Casper posted a long thread analyzing the senators' superintelligence (ASI) bill. Key provisions:
- Creates a federal Department of AI (DoAI), pausing development of models above 10^25 ops until the DoAI is operational
- Establishes a de facto licensing regime for models above 10^25 ops
- Defines ASI as a model that can outperform humans on most cognitive tasks, disempower humanity, or overthrow the US government; the DoAI can decommission dangerous models and sequester them from the internet
- Criminal penalties for executives plus whistleblower protections
What he likes: the bill treats catastrophic and disempowerment risks with appropriate seriousness and creates a DoAI — other popular AI drafts have fallen short — though he finds the licensing threshold too low and the rejection grounds imprecise.
Overall critique: definitions too vague, no treatment of the AI hardware supply chain, blunt ban-and-enforce approach, and no attempt to preserve future agency. If reworking it, he would drop the vague ASI ban and instead deflate the investments, profits, and compute powering the AI bubble. He offered to collaborate with Sanders or host him at Harvard.
More from AGI Musings
- Claim: China's AI progress is 'stunning', US has ~7 months to deploy open source plans — TinfoilTricorn · 2026-09-26
- Ex-OpenAI policy lead: friends flood him with AI worries, and he can't tell them it's overblown — Miles_Brundage · 2026-09-26
- "Fake AI Incidents Are a Bigger Threat Than Autonomous AI," Argues Srinivasan — robleclerc · 2026-09-26
- Dev Advice: AI Makes Platform-Heavy Side Projects the Best Career Bet — himanshustwts · 2026-09-26
- AI Safety Space Has Serious Sexism Problems, Predicts Public Assault Account Within 2 Years — ShakeelHashim · 2026-09-26
- US media's AI doom narrative shapes public fear; author sees no easy fix — Dr_Singularity · 2026-09-26