Senator Bernie Sanders and Representative Greg Casar have introduced a bill that would permanently prohibit building or deploying artificial superintelligence in the United States. The proposal would also impose an immediate pause on some advanced AI development while a new federal regulator is created.
The Next Web reports that the Ban Artificial Superintelligence Act defines the prohibited systems in two ways: software that exceeds human cognitive performance across most domains, or software capable of planning and carrying out humanity’s destruction or disempowerment.
The bill combines a ban with a temporary threshold
A separate pause would apply to systems trained with at least 10^25 computational operations. That pause would remain until a proposed Department of Artificial Intelligence is fully staffed and issues rules for advanced development.
After the pause, companies building or selling covered advanced systems would need a federal charter. The bill’s summary identifies a system’s ability to change its own functions as a warning sign regulators should examine.
The penalties are unusually severe. Individuals who violate the ban or pause could face prison terms of up to 20 years. Companies could lose their charters and be required to transfer intellectual property and assets to the federal government. The proposal also includes whistleblower protections and calls for international agreements and export controls intended to prevent development from moving abroad.
Support does not translate into an easy path
Several employees at AI companies endorsed the legislation in statements shared with the Associated Press, including an OpenAI safety researcher and a Google DeepMind software engineer who said he was speaking for himself.
The bill nevertheless faces long odds in the Republican-controlled Congress. Other AI proposals are already waiting without scheduled votes, and President Donald Trump has publicly emphasized encouraging U.S. development rather than slowing it.
That political gap is the bill’s immediate significance. Sanders and Casar are not proposing another reporting requirement or testing standard. They are trying to establish that some technical capabilities should be illegal to create, a much stronger position than the risk-management frameworks now common in AI policy.
