Sanders and Casar File Bill to Ban Superintelligence, With Prison Terms Up to 20 Years

Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) introduced legislation that would prohibit anyone from building artificial superintelligence, defined in the text as a system able to cause the destruction or disempowerment of humanity, including by toppling the U.S. government. The Ban Artificial Superintelligence Act attaches criminal liability to the people running frontier AI labs, with penalties reaching 20 years in prison for violations. It lands as the most punitive federal proposal yet aimed at the companies racing to train ever-larger models.

The bill's definition is the part that matters most. Rather than regulate training runs by compute thresholds or require safety evaluations, it targets the capability itself: any system that could outmatch human control, whether through economic coercion, military advantage, or direct seizure of state authority. That framing sidesteps the usual fight over benchmark scores and puts the burden on developers to prove their systems cannot cross the line. Sanders has spent years pressing tech firms on labor and antitrust issues, and Casar has pushed similar oversight themes in the House, giving the measure a two-chamber launch rather than a lone Senate gesture.

Enforcement would fall on executives, researchers, and possibly investors, not just corporate entities. A 20-year maximum puts the proposal in the same range as serious federal fraud and national security offenses, a signal that its authors view superintelligence as a category of risk closer to weapons proliferation than product safety. The bill does not yet spell out which agency would investigate, how a court would determine that a model had crossed the threshold, or whether open-weight releases would be covered. Those gaps will draw the heaviest scrutiny when the text reaches committee.

The timing is pointed. Frontier labs including OpenAI, Anthropic, Google DeepMind, and xAI have all said they are working toward systems that could exceed human performance, though they disagree sharply on timelines and on whether such systems can be controlled. Sanders and Casar are effectively asking Congress to criminalize the stated goal of the industry's best-funded players. Industry groups have argued that broad bans push research offshore and weaken the U.S. position against China, an argument that has stalled earlier safety bills.

What happens next depends on whether the bill gets a hearing at all. Sanders chairs the Senate Health, Education, Labor and Pensions Committee, which has jurisdiction over some AI workforce questions but not over model development directly, so the path to markup is unclear. The real test is whether the 20-year penalty survives contact with colleagues who favor licensing and audit regimes over outright prohibition. If it does not, the bill still shifts the Overton window: the question in Washington is no longer whether to regulate superintelligence, but whether to outlaw it.