A Senator and a Congressman Propose Banning Superintelligence Outright
On September 3, 2026, Senator Bernie Sanders (I-VT) and Representative Greg Casar (D-TX) introduced the Ban Artificial Superintelligence Act, a bill that would permanently outlaw the development and deployment of what it calls artificial superintelligence in the United States.
The bill defines the target broadly: AI systems that surpass human intelligence across the board, or systems with narrower but dangerous abilities such as the capacity to help overthrow a government or subvert their own shutdown commands. Alongside the permanent ban, the bill imposes an immediate, temporary pause on advanced AI development generally, to stay in place until a new federal regulator has written safety rules for the industry.
Sanders and Casar's own announcement frames the bill as a direct response to statements from AI lab leaders themselves, who have repeatedly said in public that today's systems are approaching or could soon reach superintelligent capability.
Twenty Years in Prison Where Brussels Writes a Check
The bill's enforcement model borrows explicitly from US nuclear weapons law, and the contrast with Europe's existing AI rulebook is the clearest way to size it up.
| Penalty for individuals | US bill: up to 20 years in prison |
|---|---|
| Penalty for companies | US bill: mandatory corporate dissolution |
| Penalty for companies (comparison) | EU AI Act: up to EUR 35 million or 7 percent of global turnover, whichever is higher |
| Enforcement body | US bill: new cabinet-level federal agency; EU AI Act: national authorities plus the EU AI Office |
| International reach | US bill: directs Washington to pursue a global ban treaty; EU AI Act: applies only inside the EU |
Where the EU AI Act, whose harshest tier already tops GDPR's ceiling, treats a violation as a very large invoice, the US bill treats it as a crime on the scale of illegal nuclear weapons work.
The Bottom Line for a Company Operating on Both Sides of the Atlantic
This bill will almost certainly not become law this session. A blanket ban on an entire technology category rarely clears a Congress where the other chamber or the White House holds a different party, and no vote has been scheduled.
What changes anyway is the argument available to regulators in Brussels. The EU AI Office began enforcing General Purpose AI obligations in August 2026, and until now its EUR 35 million ceiling was the most severe AI penalty on the table anywhere in the West. A sitting US senator now proposes prison sentences and company dissolution for the same underlying worry. Compliance teams negotiating AI Act timelines should expect that comparison to show up in Brussels hearings and in vendor risk assessments, even while the US bill itself sits in committee.
Read next: Minnesota's Ban Took Effect With the Case Still Open | A US Bill Could Switch Off the AI You Depend On
