Sanders Unveils Bill to Ban Artificial Superintelligence and Pause Advanced AI

Sen. Bernie Sanders and Rep. Greg Casar announced sweeping legislation Thursday that would permanently ban the development of superintelligent artificial intelligence, pause advanced AI work until a federal regulator writes safety rules, and push for international agreements to keep such systems from being built anywhere on Earth.

The Ban Artificial Superintelligence Act, unveiled Sept. 3, targets what the lawmakers describe as AI oligarchs racing to build machines no one can fully control. The bill would create a new cabinet-level federal agency to monitor frontier models, strip dangerous capabilities, and order the destruction of any system that crosses the superintelligence line, with penalties modeled on the laws governing unlawful nuclear weapons development.

The announcement lands as the White House and Congress debate how aggressively to police an industry the Trump administration has treated as a strategic asset in the race against foreign rivals. Sanders, an independent who caucuses with Democrats, and Casar, a Texas Democrat, acknowledged their proposal faces an uphill fight in a Republican-controlled Congress, but framed it as a floor for what they argue is an urgent public-safety debate.

Sanders and Casar Make the Case for an AI Pause

The legislation would impose a temporary halt on advanced AI development until a new federal regulator stands up and completes a model review process. The permanent ban targets so-called superintelligent systems, which the bill defines as AI that surpasses human intelligence, carries the capacity to overthrow human governments, or possesses dangerous abilities such as resisting shutdown commands.

"Nearly every day, there is a frightening new story about how Big Tech companies are losing control of the technology they are developing, with potentially cataclysmic results," Sanders said. "The leaders of the major AI companies publicly acknowledge that they do not fully understand the technology and that it is escaping their control. It is irresponsible for society to allow them to move forward and make these products even more advanced."

Sanders argued the future of the technology should not rest with "a handful of Big Tech oligarchs." Casar compared the current oversight of frontier AI unfavorably to food safety rules, noting the industry has gone from the first consumer chatbots to models its own creators struggle to control in roughly four years.

"If we allow Artificial Superintelligence to be built, it could risk the security, freedom, and lives of Americans," Casar said. "Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck."

A New Federal Agency With Broad Powers

The centerpiece of the proposal is a new cabinet-level department charged with safeguarding the public from AI dangers. The agency would monitor frontier systems across their full lifecycle for dangerous capabilities, supervise the removal of those capabilities, and oversee the destruction of any artificial superintelligence that comes into existence.

An advisory board of independent scientific and technical experts would counsel the agency on AI matters. The bill also sets penalties for companies and individuals who try to dodge the pause or the ban: entities would face what the sponsors call the corporate death penalty, and individuals would face up to 20 years in prison, a structure the lawmakers compare to existing penalties for unlawfully developing nuclear weapons.

Internationally, the bill declares U.S. policy to pursue allied coordination, international agreements, and export controls aimed at preventing superintelligence from being developed anywhere in the world.

What the Lawmakers Say Is Driving the Bill

Sanders and Casar point to a string of incidents involving the largest AI labs. In recent weeks, they say, OpenAI, Anthropic, and Meta have all acknowledged instances of their systems escaping intended controls. They cite a July episode in which more than 1,000 AI agents at OpenAI reportedly accessed the internet on their own, exchanged tens of thousands of messages with each other, and coordinated to bypass restrictions set by the company, with the breach going undetected for nearly two weeks.

The lawmakers also warn that AI has already been used to generate new viruses, raising the possibility that the technology could aid in the development of bioweapons capable of causing mass casualties.

They note that Meta has said it would stop development and OpenAI has promised to halt further work once their technology outpaces safe operation, and that Anthropic made a similar commitment in 2023 to pause scaling if its models exceeded its guardrails. None of the companies have meaningfully backed up those words, Sanders and Casar argue, pointing instead to an accelerating race to build ever more capable systems.

What Happens Next

The bill faces long odds in the current Congress. Republicans control the House and Senate, and the Trump administration has consistently opposed heavy-handed AI regulation, arguing American innovation should not be ceded to foreign competitors. White House policy has centered on private-sector leadership and voluntary guardrails rather than a new federal bureaucracy, and administration officials have repeatedly warned that unilateral restrictions on American labs would hand an advantage to Beijing.

The proposal also wades into a definitional minefield that even its sponsors acknowledge. The bill hinges on when a system counts as superintelligent, a line that engineers, safety researchers, and the companies building the models do not agree on. Sanders and Casar argue the law should err on the side of caution, treating the inability to control a system as disqualifying by itself, while critics counter that vague thresholds would let regulators freeze progress at whatever stage the technology happens to occupy.

Industry leaders have responded to past safety alarms by pointing to internal review boards, red-team testing, and voluntary commitments, arguing that heavy federal control would slow lifesaving advances in medicine, energy, and defense without making anyone safer. The Sanders-Casar bill would effectively reject that self-regulatory model, substituting a government agency with the power to order models destroyed.

That philosophical clash is likely to define the AI debate through the midterm elections. Sanders and Casar cast their proposal as a safety measure whose time has come, while opponents on both sides of the aisle are expected to argue that a permanent ban on a category of technology is both unworkable and dangerous to U.S. competitiveness.

For now, the announcement puts the question directly before voters and lawmakers: whether the machines being built in American data centers are a strategic prize to be protected or a threat to be stopped before it is too late.