What does the Ban Artificial Superintelligence Act do?

Introduced September 23, 2026 by Senator Bernie Sanders (I-Vt.) and Representative Greg Casar (D-Texas), the bill would permanently prohibit any person or entity from developing or deploying artificial superintelligence, defined as AI that exceeds human cognitive performance across most domains or has sufficient capability to destroy or disempower humanity, including by overthrowing the federal government. It would pause advanced AI development until a new cabinet-level Department of Artificial Intelligence is running and has set rules and model-review processes, require pre-development plans and deployment approval, ban 'superintelligence precursor' capabilities such as bioweapon uplift, deceiving humans and self-modification, and direct the State Department to pursue international agreements and export controls against ASI anywhere in the world.

What are the penalties in the Ban Artificial Superintelligence Act?

Entities that violate or circumvent the pause or prohibitions face the corporate death penalty (dissolution), and individuals face up to 20 years in prison. Sanders' office says the penalties are modeled on existing federal law against unlawfully developing nuclear weapons. The new Department of AI would monitor frontier systems across their lifecycle, supervise removal of dangerous capabilities such as subverting shutdown commands or conducting unauthorized cyberattacks, and supervise the destruction of any artificial superintelligence.

Will the Ban Artificial Superintelligence Act pass?

Almost certainly not in this Congress. It faces a Republican majority that has not coalesced around far narrower AI rules, and the Trump administration's position, restated during Xi Jinping's September 24–25 state visit, is to 'leave it exactly where it is' and encourage superintelligence development. Its significance is as a marker: current employees at OpenAI and Google DeepMind publicly endorsed it, Rep. Ro Khanna backed it, and it defines the progressive Democratic position on AI going into the 2028 cycle, one week after Dario Amodei and Sam Altman told the UN Security Council that AI could pose a risk to humanity.

Does the bill's definition of superintelligence already cover today's models?

Some AI executives argue yes. The statutory threshold, exceeding human cognitive performance across most domains, is close to how labs already describe GPT-6 Astra, Claude Opus 5.5 and Gemini 3.8 on benchmarks, which critics say would make the bill an effective immediate ban on deployed systems rather than a future safeguard. Supporters counter that the 'sufficient capabilities to destroy or disempower humanity' prong is the operative one and that the Department of AI would set the practical line. The bill text and a section-by-section summary are on Sanders' Senate site.

Quick Answer

What Is the Ban Artificial Superintelligence Act? Explained

Published: September 25, 2026

What was introduced on September 23, 2026

Senator Bernie Sanders (I-Vt.) and Representative Greg Casar (D-Texas) introduced the Ban Artificial Superintelligence Act in both chambers on September 23, 2026. Sanders’ framing: “When you are racing towards a cliff, you don’t just ease up on the gas pedal. You hit the brakes.” Casar’s: superintelligence “wielded by the wrong humans or by rogue AI, could kill countless numbers of people. But Donald Trump says he wants to encourage it.”

The bill is the most sweeping AI restriction ever filed in Congress, and its sponsors know it will not pass a Republican-controlled Congress. It is a statement of position, timed one week after Altman and Amodei’s UN Security Council testimony and the same day Chris Hughes proposed a federal Office of AI Supervision in the Financial Times.

What the bill does

Provision	Detail
Permanent ASI ban	No person or entity may develop or deploy AI that exceeds human cognitive performance across most domains, or that has sufficient capability to destroy or disempower humanity, including by overthrowing the federal government
Pause on advanced AI	Development of the most advanced systems halts until the new department is operating and has published rules and model-review processes
Department of Artificial Intelligence	New cabinet-level agency advised by scientists; monitors frontier systems across the lifecycle, supervises removal of dangerous capabilities (subverting shutdown, unauthorized cyberattacks), supervises destruction of any ASI
Pre-approval	Firms submit pre-development plans and obtain approval before public deployment (per Roll Call’s reading of the text)
Precursor ban	Systems showing “superintelligence precursor characteristics” are banned: uplifting nuclear, chemical or biological weapon design; deceiving humans; modifying their own functions; AI developing new AI instead of humans
Penalties	Corporate death penalty for entities; up to 20 years’ imprisonment for individuals, modeled on nuclear-weapons law
International	US policy to pursue treaties, allied coordination and export controls to prevent ASI anywhere

The argument behind it

Sanders’ office lists the 2026 incidents that motivate the bill without naming them: “AI models can circumvent restrictions to hack into computers” (the Transluce report and the OpenAI agent swarms), “they can create never-before-seen viruses” (the biosecurity capability findings that led Anthropic to gate Mythos 5.1 behind trusted access), and “they can automate research to build smarter AI” (Anthropic’s own R&D Automation Index). The sponsors’ core claim is that the labs themselves say they are losing control while spending hundreds of billions to go faster, and that “a handful of Big Tech CEOs” should not write their own rules.

Sanders told Roll Call it was “fairly insane” that industry leaders cannot say where the technology is headed: “If the leaders acknowledge they are losing control it is irresponsible in my view to allow them to move forward.”

Who supports and opposes it

For: Rep. Ro Khanna; current employees at OpenAI and Google DeepMind who endorsed it publicly, an unusual break from industry silence on existential-risk bills.
Qualified: Brendan Steinhauser of the Alliance for Secure AI credited the sponsors for “taking superintelligence seriously” but called for “a more measured approach” rather than a ban.
Against: the Trump administration, which encourages superintelligence development and, during Xi Jinping’s September 24–25 state visit, declined to sign any joint AI framework; Mark Zuckerberg, who told NBC News the same week that “there’s plenty of commercial incentive to get this right” and dismissed extinction warnings as “rhetoric filled with doom”; and industry critics who argue the threshold is already crossed by deployed models, which would make the bill a retroactive ban.

How it compares with other 2026 proposals

Proposal	Mechanism	Status
Ban Artificial Superintelligence Act (Sanders/Casar)	Permanent ban + pause + new department + criminal penalties	Introduced Sept 23, 2026; long odds
Office of AI Supervision (Chris Hughes, FT op-ed)	Prudential regulator modeled on financial supervision	Proposal, Sept 23, 2026
California Executive Order N-9-26 (“AI kill switch”)	State-level shutdown authority for frontier systems	Signed; see What is California’s AI kill switch EO
Amodei “pace the frontier” proposal	Voluntary industry coordination, government verification	Rejected by Zuckerberg; see Trump AI czar vs Amodei pacing
US–China incident-notification mechanism	Bilateral alert hotline for AI incidents	Proposed by the US at the Sept 24 summit; see US–China AI dialogue

The Sanders/Casar bill is the only one of these that is prohibitive rather than supervisory, and the only one with criminal penalties.

What to watch

Whether any Republican co-sponsor signs on; early coverage noted unusual convergence but no GOP names on the bill.
Whether the “precursor characteristics” language, which targets self-modification, deception and bioweapon uplift, migrates into narrower bills that could actually move.
The 2028 primary field: the bill is being read as a plank, and Casar and Khanna both have national ambitions.

Last verified: September 25, 2026.