Anthropic Proposed AI Safety Bill to US Congress with Restrictions on Government

Published: October 18, 2026 – 09:05 UTC

Anthropic is pushing new rules for advanced AI amid disputes over government oversight.
The company has introduced its own draft bill that limits government powers.
The bill also предусматриє independent model evaluations, mandatory risk reporting, and the rollback of some state-level legal requirements.

Anthropic has suggested US lawmakers a draft federal law on the safety of advanced artificial intelligence models that limits the government’s authority to block their release and rolls back some state-level legal requirements.

The document, reported by Politico on October 17, 2026, would allow government intervention only when specific severe risks are present, and would introduce independent model evaluations and mandatory reporting requirements for developers.

Conditions under which the Government Could Halt the Release of AI Models

According to the outlet, the draft bill was delivered to members of the US Congress. The document is meant to закрепити safety rules for advanced AI models while limiting the state’s ability to вмешатися in their development and use.

Under the proposal, the government could seek to block the release or use of a model only if it poses a foreseeable and imminent risk of more than $1 billion in damages or the death or serious injury of more than 50 people. This refers to threats related to the creation or release of biological weapons, a model bypassing a developer’s safeguards, or a cyberattack.

The draft also sets out the following requirements for developers of advanced models:

To build and maintain safety systems
To publish transparency reports on new models
To protect whistleblowers who report violations
To notify the US Department of Commerce about critical safety incidents

Independent evaluators registered with the Department of Commerce must periodically audit developers and their models. Companies will be allowed to pay for such evaluations, as long as compensation is not tied to the results. If an assessment identifies an imminent risk as defined by the bill, evaluators must notify the department within 72 hours after the evaluation is completed.

Recall that the conflict between Anthropic and the administration of US President Donald Trump escalated in early 2026 over disagreements about the use of AI in the military, after which the authorities added the company to a supply chain risk list and restricted cooperation with it by federal agencies.

Later, the US government demanded limiting foreigners’ access to the Claude Fable 5 and Mythos 5 models, prompting Anthropic to suspend support for them for all customers. At the same time, federal judge Rita Lin ruled that adding the company to the blacklist was unlawful, noting that the authorities had not proven the existence of a real threat to national security, and that the restrictions violated the First and Fifth Amendments to the US Constitution.

The Department of Commerce will be able to petition the court to halt the release of a model. At the same time, the document explicitly prohibits other federal agencies and officials from independently suspending the development, deployment, or internal use of such systems. A federal court in Washington will be able to make the final decision on halting a model.

An Anthropic spokesperson confirmed that the document is based on the company’s advanced AI safety framework.

“This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June. These are sensible policies we believe Congress should act on,” he said.

Anthropic Wants to Limit Some State Laws

The bill also предусматривает rolling back some regional requirements for public disclosure of risks, independent model evaluations, and reporting on critical incidents. The issue remains contentious in Congress: some Democrats fear that a single federal standard would weaken rules already introduced in California, Illinois, and New York. Representatives of the tech industry, by contrast, argue that differing state requirements make compliance more difficult.

Against the backdrop of the debate over the limits of government oversight, Anthropic faced sweeping export restrictions over the summer, which cut off access to two of its most advanced models. The administration’s decision raised concerns in the tech sector about the ability to restrict access to AI at officials’ discretion.

Other initiatives are being discussed in Congress in parallel. In the House of Representatives, Reps. Lori Trahan and Jay Obernolte are advancing the FRONTIER Act, which would also allow the rollout of dangerous models to be paused and would preempt some state laws. In the Senate, Majority Leader John Thune, together with Sens. Amy Klobuchar and Ted Cruz, is negotiating legislation to counter catastrophic AI risks.

At the same time, some politicians and organizations advocating for technology safety are calling for tougher government oversight, while some progressive lawmakers are urging a pause on the development of advanced AI.

As a reminder, the Trump administration also approached OpenAI with a request to limit the initial rollout of GPT-5.6 to a small circle of government-approved partners, in order to test the model ahead of a public release on national security grounds.

Сообщение Anthropic Proposed AI Safety Bill to US Congress with Restrictions on Government появились сначала на INCRYPTED.

Source: Incrypted