Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act to permanently ban artificial superintelligence in the US and create a Department of Artificial Intelligence. The bill also proposes an immediate pause on work on advanced systems trained on 10^25 or more operations until the new agency is fully staffed and has set clear rules. The proposal matters because it would replace voluntary safety commitments with criminal penalties and federal licensing for frontier development.
What the bill would ban and pause
The lawmakers introduced the bill on Wednesday and gave it first to the Associated Press, where Joey Cappelletti reported it. Sanders is an independent from Vermont, while Casar, a Texas Democrat, chairs the Congressional Progressive Caucus. The text bans building or deploying artificial superintelligence and defines it in two ways. One is a system that exceeds human cognitive performance in most domains or tasks. The other is a system able to plan and carry out the destruction or disempowerment of humanity.
The pause would start at once and end only when the new department is staffed and has set clear rules, after which any company that builds or sells advanced AI would need a charter. The summary flags a system's ability to change its own functions as a warning sign, and Semafor's Nicholas Wu reported that the text bans recursive self-improvement. Individuals who break the rules could face up to 20 years in prison. A violating company would lose its charter and transfer intellectual property and assets to the federal government.
The bill faces long odds in the Republican-controlled Congress, where three other AI bills await a vote. President Donald Trump pushed the other way on Tuesday, telling the UN the US would encourage what he called super intelligence rather than restrain it, with opponents warning a US slowdown leaves China ahead. Sanders argued the most dangerous systems require US-China cooperation, as with Cold War nuclear arms control. Several AI company employees backed the proposal, including OpenAI researcher Juan Felipe Ceron Uribe and Google DeepMind engineer Swante Scholz.
What this means for companies using AI
For buyers and builders of AI, the immediate effect would fall on the frontier-model pipeline rather than on deployed assistants and process automation. A pause above 10^25 operations and a charter requirement would concentrate compliance on a small set of model developers, with potential delays to new releases and added diligence for enterprise buyers. Large companies with direct model contracts would face new licensing checks and procurement reviews, while smaller firms using third-party APIs would feel it indirectly through slower capability gains and revised vendor terms.
The proposal remains far from law and leaves key operating terms undefined for technology decisions. The criteria for charters, the full list of warning signs beyond self-modification, and the scope of export controls on AI computing would become clear only if the department is staffed and issues rules. The news alone does not restrict current tools. Firms evaluating vendors should ask where models were trained, what compute thresholds apply, how self-modification is prevented, and what contractual cover exists if releases are delayed by regulation.
Whether the approach gains traction can be judged by congressional action after recess and by the 2028 presidential debate. Sanders said Congress must act as soon as it returns and wants candidates to take up the issue now. Movement on the three waiting AI bills, a hearing date, or draft charter criteria would show the debate shifting from statements to operating constraints for the AI supply chain.
