Senator Bernie Sanders and Representative Greg Casar have introduced a bill that would flat out ban the development of AI that surpasses human capabilities. Companies that break it would face up to 20 years in prison for individuals, and dissolution for the company itself.
Key Takeaways
- The Ban Artificial Superintelligence Act bans any AI that would exceed human cognitive performance across most domains
- The bill creates a federal Department of Artificial Intelligence tasked with writing safety rules before advanced development can resume
- Penalties reach 20 years in prison, the same sentencing bracket used for illegally building nuclear weapons
Have an AI Sum Up This Article
ChatGPTBernie Sanders’ Bill Goes Further Than Anything Before It
The Ban Artificial Superintelligence Act, unveiled on September 23 in a joint statement from Bernie Sanders and Greg Casar, crosses a line most bills in Congress have avoided so far. Earlier proposals targeted specific uses, deepfakes, facial recognition, autonomous weapons. This one goes after the broadest category available: any AI whose cognitive capabilities would exceed humans across most domains, or that could threaten or overthrow a government outright.
The bill does two things at once. It permanently bans the development and deployment of artificial superintelligence. And it temporarily pauses advanced AI development in general, until a new Cabinet-level Department of Artificial Intelligence sets safety rules and a supervision framework. Sanders also wants Washington to push this framework internationally, to keep superintelligence from emerging elsewhere in the world even if the US manages to contain it at home.
The timing is no coincidence. The bill lands days after Claude’s own founder, Dario Amodei, called for slowing the industry’s pace, and right as Anthropic revealed it had left a bioweapon safety filter switched off for 11 months without noticing. Sanders cites both episodes directly to justify the urgency behind his bill.
20 Years in Prison, the Nuclear Weapons Bracket
The penalty structure shows how far this bill is willing to go. A company that develops or deploys superintelligence would face what Sanders calls the “corporate death penalty,” meaning forced dissolution. An individual found responsible would face up to 20 years in prison, the exact sentence reserved in the US for illegally building nuclear weapons. The comparison is deliberate: for Sanders, building superintelligence without guardrails belongs in the same danger category.
That severity cuts against the approach other lawmakers have taken. Around the same time, OpenAI asked Congress whether voluntarily slowing AI development would even be legal under antitrust law, a question that assumes no binding rule exists yet. The Sanders-Casar bill would close that gap in one move, leaving companies no room to slow down or speed up on their own terms.
The bill splits opinion even among researchers who take AI safety seriously. Gary Marcus, a researcher known for his repeated criticism of the industry, published a public objection to the bill, arguing the definition of “superintelligence” is too vague to enforce without creating massive legal uncertainty for the entire industry.
More articles on Horizon
- OpenAI Hacked an Australian Government Website
- ChatGPT Voice Now Reads Your Email and Calendar
- OpenClaw: Meta Admits It Copied the Open Source Agent
A Republican Congress Stands in the Way
Politically, the bill starts at a serious disadvantage. Congress remains under Republican control, and lawmakers there have struggled to agree on rules far narrower than this one. Sanders knows it: he is framing this bill less as legislation bound to pass and more as a marker, a way to publicly draw the line before the industry crosses it for good.
For anyone using AI tools day to day, nothing changes right away. Claude, ChatGPT, and Gemini will keep running exactly as they do now, and none of them comes close to the definition the bill sets. The stakes lie further out, in the path the next generation of models actually takes.
For the labs themselves, the signal cuts differently. Dario Amodei recently traced the growing backlash against AI back to a lack of transparency from the industry itself, and a bill like this one is the direct consequence: when companies fail to convince lawmakers they can regulate themselves, elected officials eventually step in with tools far blunter than the industry would have picked on its own.
Follow the story on Horizon.
