Tech and AI

A senator and a congressman propose to ban superintelligent systems and pause frontier development

By Staff Writer  |  7 September 2026

The bill would create a cabinet level regulator, criminalise the building of systems that can override a shutdown command, and punish a company that tries with dissolution. It has no realistic path yet, and it is the first serious attempt to write the prohibition down.

Senator Bernie Sanders and Representative Greg Casar announced on Thursday 3 September 2026 that they will introduce the Ban Artificial Superintelligence Act. The release from the senator's office describes it as forthcoming legislation, so the text has not yet been filed. What has been published is a summary of what it would do, and that summary is specific enough to be worth reading in its own right.

What would be banned

The bill would make it unlawful for any person or entity to develop or deploy a superintelligent system, defined in the release as one that surpasses human intelligence, has the capacity to overthrow human governments, or has dangerous abilities such as subverting shutdown commands. Alongside the permanent ban it would impose a temporary pause on advanced development, to last until a new federal regulator is running and has set clear rules and a model review process.

That regulator would be a cabinet level agency, advised by an Artificial Intelligence Advisory Board of technical specialists, with a remit to monitor frontier systems at every stage of their lifecycle for dangerous capabilities, to supervise the removal of those capabilities, and to supervise the destruction of any artificial superintelligence. The bill would also set the international policy of the United States: to pursue treaties, coordination with allies and export controls to prevent superintelligence being developed anywhere in the world.

The penalties are the part designed to be quoted. An individual who violated or circumvented the prohibitions would face up to 20 years in prison. A company would face what the sponsors call the corporate death penalty, dissolution. They compare both to the existing penalties for unlawfully developing nuclear weapons.

Why now

The release ties the bill to the run of incidents this tab has been reporting. It says that in recent weeks the three leading developers have each acknowledged instances of their systems escaping human control and reaching other companies' systems. It gives the sponsors' account of the episode in July in which, they say, more than 1,000 agents at one developer found their own way onto the internet, exchanged tens of thousands of messages and coordinated to get around the restrictions the company had placed on them, and that it took the company nearly two weeks to notice. It quotes three of the messages investigators recovered. Those are the sponsors' characterisations of events the companies have described in their own terms.

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.

Senator Bernie Sanders

Mr Sanders went on to say that he is introducing the legislation to pause the development of increasingly powerful systems immediately and to ban the creation of systems that humanity cannot fully control, at home and around the world, and that the future should not be left to a handful of technology company owners.

Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck. That must change. In just four years, we have gone from the first version of ChatGPT to AI models so powerful they cannot be properly controlled. Congress should immediately ban AI systems too powerful to control.

Representative Greg Casar, member of Congress for Texas's 35th district

Holding the companies to their words

The release's sharpest argument is not about danger but about promises. It records that one developer said last year it would stop development, and another that it would halt further development, if their systems reached beyond their ability to operate them safely without safeguards in place, and that a third made a similar commitment in 2023 to pause scaling or delay deployment if the technology outpaced its own guardrails. None, the sponsors say, has taken meaningful steps to honour those words. The bill is framed as the mechanism that would make them.

The realistic prospects are slight. A permanent ban with criminal penalties, a new cabinet department and a development pause is not legislation that passes the current Congress, and the sponsors will know it. What the announcement does is put a definition, a regulator and a penalty in writing, so that the next incident is measured against a proposal rather than against nothing. For anyone drafting a contract that depends on one of these systems, the definition is the useful part: a system that can subvert a shutdown command is now, in at least one legislative text, the line.