Sanders Bill Would Ban 'Superintelligence', 20 Years Prison

Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) have introduced legislation that would ban anyone in the United States from developing artificial superintelligence, with violators facing up to 20 years in prison.

The bill, called the Ban Artificial Superintelligence Act, defines the targeted technology as AI capable of the "destruction or disempowerment of humanity," including by overthrowing the government. That language puts the legislation's stakes in unusually stark terms for a congressional proposal. The 20-year prison sentence would apply to AI industry leaders who violate the law.

The measure reaches beyond superintelligence alone. It would also pause the development of advanced AI systems more broadly, according to the lawmakers' announcement — a provision that would put the bill in direct conflict with the current direction of federal AI policy.

The timing matters. President Donald Trump said the US will use the term "superintelligence" in place of "AI" in official documents, signaling how central the concept has become to the administration's posture on the technology. A legislative attempt to criminalize its development lands in the middle of an escalating debate over how fast advanced AI should be built and who should control it.

The proposal, introduced via Sanders' Senate office, would also create a new federal agency to enforce the ban, according to the press release announcing the legislation.

For AI companies racing toward more capable systems, the bill signals a growing willingness among some lawmakers to treat existential AI risk as a criminal enforcement matter rather than a voluntary-commitments issue. For policy watchers, it marks one of the most aggressive legislative interventions proposed to date in the US, contrasting sharply with an administration that has framed AI dominance as a national priority.

The bill faces long odds in a Congress where AI legislation has moved slowly and along partisan lines. But its introduction reframes the range of options on the table: if even a fraction of its enforcement logic survives into future compromise legislation, AI executives could eventually operate under criminal, not just regulatory, liability for the systems they build.