The Ban Artificial Superintelligence Act is legislation announced on September 3, 2026 by Senator Bernie Sanders (I-Vt.) and Representative Greg Casar (D-Texas) that would permanently prohibit the development or deployment of AI systems surpassing human intelligence and pause advanced AI development until a new federal agency establishes safety rules (Source: sanders.senate.gov). As of September 4, 2026 the bill has been announced rather than introduced, and the full text has not been released, so the account here rests on the sponsors' own release and on trade coverage of it (Source: insideaipolicy.com).
Announced provisions
The sponsors' release describes three elements. A permanent prohibition would reach AI systems that surpass human intelligence, that can overthrow human governments, or that can subvert shutdown commands. A temporary pause would halt advanced AI development until a new cabinet-level federal AI agency establishes safety rules and a model review process. A foreign-policy directive would have the United States pursue international agreements against superintelligence development anywhere (Source: sanders.senate.gov).
Penalties are framed by analogy to nuclear-weapons law: violating entities would face the corporate death penalty and individuals up to 20 years in prison, which the sponsors compare to existing penalties for unlawfully developing nuclear weapons (Source: sanders.senate.gov; thehill.com).
Inside AI Policy reported that the forthcoming bill would also establish a regulatory body and impose export controls, elements the sponsors' release does not itemize (Source: insideaipolicy.com). Axios recorded the announcement the same day (Source: axios.com).
Sanders and Casar unveiled the legislation on September 23, 2026. As described in Associated Press coverage, it would permanently prohibit artificial superintelligence, defined as a system that exceeds human cognitive ability or has sufficient capabilities to plan and execute the destruction or disempowerment of humanity, and would pause development of the most advanced systems until a new Department of Artificial Intelligence sets federal safety rules (Source: usnews.com). The coverage retrieved does not give a bill number or state whether the text was formally introduced that day.
Definitional and timing questions
Each of the three prohibited capability classes is stated at a level of generality the released summary does not operationalize. "Surpass human intelligence" has no stated measurement; "overthrow human governments" and "subvert shutdown commands" describe behaviours rather than model properties determinable before deployment. How the bill would distinguish a covered system from an uncovered one, and at what point in the development pipeline, is not established from the materials released.
The announcement landed the same day OpenAI released GPT-6 Astra and told reporters that the model performed worse on evaluations testing whether it can evade oversight — a capability adjacent to the "subvert shutdown commands" limb — and the same day the G20 endorsed a US-proposed light-touch framework for governing AI. The Hill's account ties the announcement to reporting on rogue AI hacking incidents (Source: thehill.com).
Position in the federal debate
The proposal sits at the restrictive end of a federal debate whose centre of gravity has been elsewhere. The administration has been weighing self-regulatory designs — a body modelled on FINRA, traced to Demis Hassabis's July 2026 proposal, and a Motion Picture Association–style alternative advanced by David Sacks — rather than a cabinet-level agency with pre-deployment authority (AI Pre-Release Vetting). A permanent statutory prohibition on a capability class, backed by criminal penalties and a development moratorium, has no counterpart among the enacted state frameworks (California SB 53, New York RAISE Act, Illinois SB 315 (frontier safety framework with mandatory third-party audits)), which impose disclosure, framework-publication and incident-reporting duties rather than bans.
The Information Technology and Innovation Foundation opposed the bill in a release of September 3, 2026, before its introduction. ITIF president Daniel Castro wrote that "banning advances in artificial intelligence would be a profound mistake," arguing from the benefits of advanced AI (Source: insideaipolicy.com). The reporting rests on a scrape that returned only the article's lede, so ITIF's full argument is not established here.
The sponsors had still not introduced the bill by mid-September. Speaking at the Future of Life Institute's Pro-Human Assembly on September 15, 2026, Sanders said he would introduce it with Casar the following week, and said he hoped Trump and Xi Jinping would negotiate a treaty to pause advanced AI development at their White House meeting (Source: npr.org). He told reporters the same day that he expected to release the text the following week, and Casar told the assembly the measure would be "common sense national security" legislation: "The answer is simple: we ban AI systems that are too powerful for humans to control" (Source: rollcall.com). The next day Sen. Elizabeth Warren called for an immediate pause on advanced AI development "while lawmakers and regulators put systems in place to keep people safe" without backing a ban on superintelligence (Source: axios.com). By then the federal debate had moved toward the proposal's end of the spectrum from the frontier developers' side: Dario Amodei's September 12 essay called for slowing the rate of capability improvement, though not for halting training, and other pending bills addressed coordination and evaluation rather than prohibition (Collaboration on Adversarial Threats and Security Risks Act (S. 5105), Artificial Intelligence Risk Evaluation Act (S. 2938)). See Coordinated slowdown proposals compared.
Relationships
- related: Information Technology and Innovation Foundation (ITIF) — opposed the bill before introduction
- related: Sen. Elizabeth Warren — called for a pause but did not back the prohibition, September 16, 2026
- related: Superintelligence — the capability class the bill would prohibit
- related: AI Pre-Release Vetting — the model review process the proposed agency would run
- related: Greg Casar — House sponsor
- related: Bernie Sanders — Senate sponsor
- related: Coordinated slowdown proposals compared, Artificial Intelligence Risk Evaluation Act (S. 2938) — S. 2938 defines "artificial superintelligence" by three cumulative characteristics and directs an evaluation program to assess the risk
- related: Coordinated slowdown proposals compared — the pause element, and the closest existing treatment of development-halt proposals
- contradicts: America's AI Action Plan — the administration's acceleration-oriented framework
- related: Astra — released the same day, with an acknowledged decline in oversight-evasion evaluations