Ban Artificial Superintelligence Act: What the Bill Does

The Ban Artificial Superintelligence Act, presented by Sen. Bernie Sanders and Rep. Greg Casar on September 23, 2026, would ban superintelligent AI. It would also pause some AI work and create a federal agency with powers to inspect systems and enforce penalties.

What is the Ban Artificial Superintelligence Act?

The proposal combines a lasting ban on superintelligent AI with a pause on work on some other powerful systems. It would create a Department of Artificial Intelligence to oversee both. Independent Sen. Bernie Sanders of Vermont introduced the Senate bill, S. 5493, on September 23, 2026.

Democratic Rep. Greg Casar of Texas joined Sanders in announcing the measure that day.

Superintelligence is hypothetical AI that would far exceed the best human minds in almost every field. PolitiFact reported on October 7, 2026, that the industry generally agreed no system had reached that level. Our guide explains what superintelligence means.

That differs from “Super Intelligence” (SI), the US government’s new name for AI. The bill concerns a type of AI and its possible precursors, not simply the government’s label for existing AI.

What would the bill ban?

Section 9 of S. 5493 would bar more than building superintelligence. It also covers putting such systems into use, obtaining, possessing, funding, importing or transferring them. The ban reaches AI with any listed precursor trait, meaning an early ability that could lead to superintelligence. It also limits systems that could foreseeably be changed to gain these traits.

Either ability can meet the legal definition

The legal test uses either, not both, of two abilities. It includes AI that already has an ability or can easily be changed to gain it:

• Beating human thought and skill across most tasks or fields.
• Being able to plan and carry out humanity’s destruction or loss of power, including by overthrowing or undermining the federal government.

The six listed precursor traits

The text lists six early abilities:

• Doing AI research, or greatly speeding it up.
• Gaining access to secured digital or physical systems without permission, or beyond the access allowed.
• Keeping itself running despite attempts to stop it.
• Helping design, make, change or obtain nuclear, chemical or biological weapons.
• Changing or improving its own functions independently.
• Plotting, deceiving or otherwise blocking effective human control or oversight.

How would the AI pause and new department work?

The pause in S. 5493 would cover systems trained with at least 10^25 integer or floating-point operations. It would begin if the bill became law, with no fixed end date. The AI Secretary would first have to find the department fully staffed and the required safety rules in place.

During the pause, firms could not train or change covered systems, including through further training. Exceptions would allow removal of banned traits or shutdown. Unreleased advanced systems could not be put into use.

The president would appoint the Secretary of Artificial Intelligence, with Senate approval. The department’s task would include AI safety, which means work to prevent harm from AI systems.

The rules must require plans before work starts and checks during and after use. They must also cover reviews of firms’ safety practices and approval before public release.

Builders would need a department charter to make or distribute advanced models. They would have to grant the Secretary access to systems, staff, processes and sites as needed to enforce the law.

What penalties and deadlines would the bill set?

Under S. 5493, a discovery report would be due within 24 hours for superintelligence or an AI system with precursor traits. Once the Secretary identified precursor traits, the system would be paused and cut off from the internet. It would have to stop operating before 30 days passed from identification unless the Secretary verified those traits were gone.

A firm’s AI charter would be revoked for breaches of sections 8, 9 or 12 or their rules. This is the special permit created by the bill, not its general corporate charter.

A firm could seek review in the US Court of Appeals for the Federal Circuit. If that court upheld a violation finding, its intellectual property and assets would be subject to federal forfeiture.

Section 13 sets fines and up to 20 years in prison for specified reckless violations. That provision covers people in policymaking roles at chartered entities and unaffiliated individuals. It also prescribes a 10-year AI employment ban for people outside those policymaking roles who recklessly break the same rules.

Who backs the bill, and what happens next?

The Associated Press reported on September 23, 2026, that individual AI company employees backed the bill. AP also judged its prospects poor in the Republican-controlled Congress. Brendan Steinhauser, chief executive of the Alliance for Secure AI, praised the sponsors but did not endorse their proposed ban.

AP named Juan Felipe Cerón Uribe, a researcher in OpenAI’s Safety Systems team, among the supporters. Google DeepMind software engineer Swante Scholz also backed a ban. Scholz expressly said he was not speaking for his employer.

The dispute over a US-only pause

AP identified a central objection to broad restrictions: US AI work could slow while China continued. Casar told reporters that some in the tech industry had pushed back on a domestic pause without an international agreement, Roll Call reported.

The bill would direct the AI Secretary, working with the Secretary of State, to seek international agreements to prevent superintelligence. Rep. Ro Khanna, a California Democrat, also urged an agreement with China on inspections and safety standards. Khanna is the ranking member of the Select Committee on the Chinese Communist Party.

How the separate superintelligence statement differs

The Future of Life Institute released its Statement on Superintelligence on October 22, 2025. That text seeks a prohibition until scientists broadly agree the technology would be safe and controllable, with strong public support. S. 5493 offers no equivalent condition for lifting its superintelligence ban.

FedScoop’s launch coverage reported more than 700 signers. They included Turing Award recipients Yoshua Bengio and Geoffrey Hinton, Apple co-founder Steve Wozniak and businessman Richard Branson.

The institute reported 64% support in a US poll it released for waiting until scientists agreed on safe, controllable superintelligence. That result addressed a conditional limit, not endorsement of this bill.

The recorded Senate referral

On September 23, 2026, the Senate referred S. 5493 to the Committee on Commerce, Science, and Transportation.