Ban Artificial Superintelligence Act
- Introduced
- Passed House
- Passed Senate
- Became Law
Introduced in both chambers; in committee
Would create a cabinet Department of Artificial Intelligence that licenses who may build or distribute large AI models, freezes frontier development, and can cut systems off from the internet and destroy them.
Latest update
Rep. Alexandria Ocasio-Cortez (D-NY) signed on as a cosponsor of H.R. 10538
Take action
This bill is still moving, so there's still time to push back.
Contact CongressThe bill
Read S.5493S.5493 official recordRead H.R.10538H.R.10538 official recordKey facts
- Where
- United States
- Bill number
- S.5493 / H.R.10538 (119th)
- Status
- Introduced
- Introduced
- S.5493: September 23, 2026
H.R.10538: September 24, 2026 - Threat
- High
- Last update
- September 28, 2026
Topics
Share this bill
What it does
The Ban Artificial Superintelligence Act (S. 5493 from Sen. Bernie Sanders, H.R. 10538 from Rep. Greg Casar) creates a Department of Artificial Intelligence run by a presidentially appointed Secretary. No person may develop or distribute an "advanced" AI model (trained with 10^25 or more operations, a threshold the Secretary adjusts yearly) without a charter from the Secretary. To get and keep a charter, a developer must give the Secretary full access to its systems, processes, staff and physical facilities, and must agree to forfeit its intellectual property and assets if the charter is revoked for a violation and the revocation is not overturned on appeal. From enactment, all advanced systems are under a "mandatory pause" (no training, modification or fine-tuning) and no unreleased advanced system may be deployed until the Secretary decides the Department is fully staffed and has written its rules, with no deadline. Those rules must include filing development plans in advance, ongoing monitoring and audits, and final government approval before any advanced system is released to the public. Separately, for AI systems of any size, the bill bans developing, possessing, funding or transferring anything with "superintelligence precursor characteristics" (a list that includes the capacity to accelerate AI research, gain unauthorized network access, help design or obtain nuclear, chemical or biological weapons, or modify its own functions), and bans releasing any system that "may be foreseeably modified" to gain them. The Secretary must cut flagged systems off from the internet at once and render them inoperative within 30 days unless the Secretary verifies the characteristics are gone. Anyone who discovers such a system must notify the Secretary within 24 hours. Reckless violations carry up to 20 years in prison for company decision-makers and for "rogue actors" (anyone not affiliated with a charter holder); other employees face a 10-year ban from the AI industry. The bill also adds whistleblower protections and makes it US policy to stop superintelligence development worldwide, which may include export controls on AI computing infrastructure.
What’s at stake
The bill turns the building and publishing of large AI models into a government-licensed activity, with an open-ended freeze until a new agency finishes its own rules. Its broadest bans reach beyond the licensed frontier labs: the "foreseeably modified" and "precursor" tests are vague enough to cover open-weight releases of almost any capable model, since anyone can fine-tune published weights, and independent developers outside the charter system are the "rogue actors" who face prison. One appointed official decides what counts as dangerous, inspects company premises and staff, and can order systems offline and destroyed, and the only court review the bill provides is a Federal Circuit appeal of a charter revocation. Publishing code and model weights is a form of expression, and this bill puts prior government approval in front of it.
Our take
This is a federal licensing regime for software, with prison terms for people who build outside it. The charter covers the largest models. The ban on releasing anything that "may be foreseeably modified" into something dangerous has no size limit, so it reaches open models of any size. Whatever one thinks of frontier AI risk, a single official with power to inspect, freeze, disconnect and destroy systems, backed by forfeiture of a company's intellectual property, is a model of state control that will not stay confined to AI.
Timeline
Rep. Alexandria Ocasio-Cortez (D-NY) signed on as a cosponsor of H.R. 10538Latest
Reclaim The Net published the bill text and its analysis of the charter, pause and shutdown powers
House version, H.R. 10538, introduced by Rep. Greg Casar with nine original cosponsors; referred to the House Oversight and Government Reform Committee
Introduced in the Senate as S. 5493; referred to the Senate Commerce, Science, and Transportation Committee
The record
Sponsor
Bernie Sanders (I-VT)
Committee
Senate Commerce, Science, and Transportation Committee, House Oversight and Government Reform Committee
Suggest a bill or correction
Know a bill we should track, or spot something out of date? Email the bill name, the jurisdiction and a link to an official source.