What the bill does
Senator Bernie Sanders of Vermont and Representative Greg Casar of Texas introduced the Ban Artificial Superintelligence Act on 23 September. It is the most restrictive AI bill yet put before Congress, and it does three things at once.
It permanently prohibits the development and deployment of artificial superintelligence, which the bill defines as an AI that exceeds human cognitive performance across most domains, or that has sufficient capability to destroy or disempower humanity, including by overthrowing the federal government.
It temporarily pauses advanced AI development, though not all AI development, until a federal regulator exists and has written safety rules. And it creates that regulator: a cabinet-level Department of Artificial Intelligence with a scientific advisory board, which would review pre-development plans and sign off before a system reaches the public.
Precursor characteristics, and the penalties
The pause does not turn on a compute threshold, which is how most frontier-AI rules have been drafted so far. It turns on capability. The bill bars systems with what it calls superintelligence precursor characteristics, a list that includes the capacity to enhance weapons, to deceive humans, and to modify itself.
The penalties are unusual in technology legislation. Entities that violate the ban would face what the sponsors describe as a corporate death penalty. Individuals would face up to twenty years in prison, a term the sponsors compare to the penalties attached to nuclear weapons offences.
The bill also directs the United States to pursue international agreements and export controls so that the prohibition does not simply move development offshore.
Quotes, and the arithmetic
“Our bill bans the development of artificial superintelligence and pushes for international agreements so that no one, anywhere, builds AI too powerful for humans to control,” Casar said.
Sanders framed it as a failure of voluntary governance. “When the future of humanity is at stake, we need binding international safety rules, not voluntary standards from the industry,” he said. He has also pointed at the labs’ own statements: if the people running these companies acknowledge they are losing control, he argued, it is irresponsible to let them keep going.
Casar was careful to say the bill would not pause all AI systems, and spoke favourably about medical applications.
Its prospects are poor. The bill has no number yet, no Republican co-sponsor, and arrives while the administration is moving in the opposite direction. Even parts of the AI safety movement have declined to endorse it: the Alliance for Secure AI praised the sponsors’ intent but called for a more measured approach rather than a ban.
What the bill does is set a marker. Everything Congress has seriously considered so far — disclosure regimes, incident reporting, state pre-emption fights — regulates how frontier systems are built. This is the first federal bill to say that a category of system should not be built at all, and to attach a prison term to building it. The number to watch is not the vote count, which is unlikely to be tested, but the co-sponsor list.
