A leading AI development company is advocating for a federal safety bill that would allow the government to pause its AI models only in the face of imminent disaster.
A draft of this bill, named the Artificial Intelligence Evaluation, Governance, and Incident Safety (AEGIS) Act, suggests that the Commerce Department would seek a court order to intervene if an AI model poses a risk of causing fatalities or significant injuries to over 50 individuals, or economic damages exceeding $1 billion. Anthony Cimino, who leads federal affairs at Anthropic, expressed his defense of these parameters on social media.
The draft states, “Enumerated risk” refers to a considerable and foreseeable danger resulting from the development, upkeep, or deployment of an AI model that could contribute to the death of, or serious injury to, 50 or more people, or to property damage exceeding $1 billion.” On a recent Sunday, Cimino emphasized that these thresholds were directly adapted from existing California standards.
He also pointed out that this does not absolve companies from liability, nor does it hinder the government’s ability to ensure public safety. When approached for comment, Anthropic directed inquiries to Cimino’s post.
I lead our federal affairs team at @AnthropicAI and wanted to share a few thoughts on the draft federal AI safety bill text that’s been circulating this weekend after a version was shared with the press. First, some process. We’ve been advocating for the concepts in our Advanced AI Framework proposal for months. Understandably, federal policymakers wanted to know what this could look like in bill form. This draft is a translation of those ideas into legislative language, shared with a small number of congressional offices that have been working on these issues. It doesn’t include everything we would ultimately like to see passed into law. In part because some ideas are so new that we are still working on if or how they can be converted into policy language.
— anthony cimino (@ajcimino) October 18, 2026
The bill has not yet been officially introduced in Congress, and no sponsors are attached to it as of now.
According to the draft, the government could initiate a pause if a model aids in creating biological weapons, executing cyberattacks, escapes user control, or engages in research that heightens associated risks. These risks also feature in a blog post by Anthropic regarding its “Policy on the AI Exponential,” which ties back to the company’s June frameworks.
For a model to be paused, the commerce secretary would need to give written consent, alongside a court order from the U.S. District Court for the District of Columbia. However, the draft specifies that the government wouldn’t have the authority to destroy an AI model or its associated data.
Charlie’s analysis is right but he is nicer than I am – I found myself viscerally angry at multiple points reading this bill. There are some good and sensible parts, but overall this legislation is not remotely close to meeting the moment and I think overall passing it would make us less protected against catastrophic risks than we are under existing law. There are two notable concerns: 1. Some of the preemption provisions are remarkably far-reaching and consequential, and would remove general powers that state courts have to protect the public. Most outrageously, it has a provision that seems to stop states from enforcing laws that would protect citizens from imminent catastrophic harms. Such powers to block model deployments would only rest with the federal government. 2. Moreover, the federal government’s ability to restrict AI model development faces overly high thresholds. A catastrophic risk must be imminent, but that’s not all; the agency head must demonstrate there are no alternatives to prevent the catastrophe. This is unreasonable.
— Nathan Calvin (@_NathanCalvin) October 17, 2026
States, as stated in the draft, would not be allowed to enforce any laws that place obligations on AI developers regarding the safety protocols covered in the bill.
In a post on social media, Calvin expressed his frustration, noting that while some aspects of the bill are reasonable, overall, it does not adequately address current needs.
The proposed thresholds in the bill align with California’s Transparency in Frontier Artificial Intelligence Act, which was passed in 2025 and previously backed by Anthropic.
