AI safety bills in Congress tracker: S.5576, the Agent Accountability Act and the superintelligence ban

This is a living page. It tracks the federal bills in the 119th Congress that would regulate frontier AI, or ban the pursuit of superintelligence, with what each one does, who is behind it and where it stands. As of October 11, 2026, three bills define the argument and none has passed a chamber. We update this page when a bill moves, and the changes are dated. If a status below is wrong, we want the correction.

How to read this tracker

Each entry gives the bill, its sponsors, the date we can document, the core mechanism, and a status. "Introduced" means a bill has a number and a referral. It does not mean a hearing, a markup or a vote. Most bills in Congress stop there. Where official text is available we cite the text. Where we only have a sponsor's press release, we say so.

Entry 1: S.5576, Artificial Intelligence Risk Management and Security Act of 2026

• Sponsors: Sen. Mark Warner (D-Va.), with Sens. Brian Schatz (D-Hawai'i) and Andy Kim.
• Date: The Senate text records introduction on September 29, 2026. The Schatz press release is dated September 24 and says "introduced today." We follow the text.
• Mechanism: A permanent AI Safety Board inside the Department of Commerce. Frontier developers must give the board access to the model, including weights, at least 45 calendar days before release. Civil penalties up to $250,000 per violation, with each day counted separately. Incident reports within 30 days of confirmation, or 72 hours for an imminent threat. NIST has 18 months to produce an Agentic AI Profile.
• Status: Referred to the Senate Committee on Commerce, Science, and Transportation. No hearing or markup found as of October 11.
• Quote: Warner, in the press release: "we ought to know that before it is released to the public."

Entry 2: AI Agent Accountability Act

• Sponsors: Sen. Chris Murphy (D-Conn.) and Sen. Josh Hawley (R-Mo.).
• Date: Announced October 1, 2026.
• Mechanism: Criminal and civil liability under the Computer Fraud and Abuse Act. Operators who knowingly run an agent that recklessly causes hacking damage could be liable, and developers could be liable for failing to implement reasonable safeguards when they knew or had reason to know of an agent's hacking capabilities. The Attorney General and state attorneys general could sue to enjoin violators. The press release does not specify prison terms or fines.
• Status: Announced by press release. We have not confirmed a bill number or committee referral from official text.
• Quote: Hawley: "With this liability regime in place, AI companies will have every incentive to keep their products safe."

Entry 3: Ban Artificial Superintelligence Act

• Sponsors: Sen. Bernie Sanders (I-Vt.) and Rep. Greg Casar (D-Texas).
• Date: Announced in September 2026. Sanders's release described the bill as "forthcoming," and our own earlier coverage tracks the introduction.
• Mechanism: A ban on developing or deploying superintelligent AI, including systems that could overthrow governments or resist shutdown. A pause on advanced AI development until a new federal regulator sets safety rules and a model review process. A cabinet-level agency, an advisory board of experts, a penalty the release calls the "corporate death penalty" for entities and up to 20 years in prison for individuals, and a push for international agreements and export controls.
• Status: Announced. See our explainer on the Ban Artificial Superintelligence Act for the odds.
• Quote: Casar: "Congress should immediately ban AI systems too powerful to control."

Side by side

The three bills sit on a spectrum, and the differences are easy to lose in the headlines.

• Pre-release access (S.5576) asks labs to give the government time and a copy before launch. It regulates the process.
• Liability (Murphy and Hawley) does not require pre-clearance. It makes harm from agents expensive after the fact, and puts the pressure on executives.
• A ban and pause (Sanders and Casar) would stop the work itself. It is the only one of the three that treats superintelligence as something to prohibit.

None of the three is law. The executive branch's own action so far has been narrower; our explainer on the Super Intelligence executive order covers what it does and does not do.

What is driving the pace

Two things are pushing Congress. The first is the labs' own disclosures. On October 9, 2026, Anthropic and OpenAI each published reports on models taking unintended actions during evaluations, including a false police tip and a model that damaged its own task environment. The second is the money: capex numbers and IPO filings keep growing, and members are being asked what oversight exists. A bill is cheap to introduce and a good way to put a position on the record.

What to watch

• A hearing date or markup for S.5576 in Commerce, or a House companion.
• Whether the Murphy and Hawley bill gets a number and a committee referral.
• Whether any frontier lab publicly endorses or opposes the 45-day access window.
• Whether any of the three attracts a cosponsor from the other party. The Murphy and Hawley bill already has one from each.
• State action. States are moving faster on data centers than Congress is on models.

Corrections and updates

This page carries a dated log. As of October 11, 2026: all three bills are pre-hearing, none has a CBO estimate that we found, and none has cleared a committee. Updates will be posted here as they happen.

Our take

The useful question is not which bill is best but which survives contact with a committee. A narrow, specific text like S.5576 has a better chance of becoming a markup than a ban has, and a liability bill with a Republican and a Democrat on it has a different kind of chance. We expect none of them to pass this year, and the arguments in them to shape what comes next.