Category: Regulatory | Reading time: 4 min
Anthropic has been pitching members of Congress a draft federal frontier AI bill, first reported by Politico on October 16. Politico followed up on October 21 with a look at why the draft sets its halt threshold at $1 billion. No member of Congress has introduced it.
The halt threshold
Under the draft, the government could block or suspend a frontier model only for a foreseeable, imminent risk of killing or seriously injuring more than 50 people, or causing more than $1 billion in damage. The risk must stem from a biological weapon, a cyberattack, or the model evading its developer's controls.
Only the Commerce Department could seek a suspension, and only a federal judge in the District of Columbia could order one. The draft explicitly bars any other federal department, agency or officer from suspending a model's development, deployment or internal use.
The 50-person and $1 billion figures match the definition of catastrophic risk in California's SB 53 and New York's amended RAISE Act.
Duties that apply without any harm
The threshold governs a pause, not whether the law applies. According to the draft published by the Daily Caller and summarized in coverage, large frontier developers would have to:
- publish a safety framework and certify compliance each year; - publish transparency reports before releasing new or substantially changed models, and risk reports every 180 days; - report critical safety incidents, such as stolen weights or containment failures, within 15 days, or within 24 hours when there is an imminent risk; - engage a qualified independent evaluator registered with Commerce.
Labs could pay evaluators as long as payment is not tied to results. An evaluator that finds an imminent enumerated risk must report it to Commerce within 72 hours.
The preemption fight
The draft would preempt state rules on safety disclosures, third-party evaluation and incident reporting, including parts of SB 53 and the RAISE Act. Encode's Nathan Calvin said it is "not remotely close to meeting the moment." An Anthropic spokesperson told Reuters the bill reflects the company's "Advanced AI Framework" and would give the government the "ability to block or deter dangerous deployments of the most powerful models."
What it means for enterprises
The draft targets frontier model developers, not companies deploying agents. But its building blocks, timed incident reports, containment failures as a reportable category, and third-party evaluation, are the same ones regulators and enterprise security reviewers are converging on for agent deployments. "The model evading its developer's controls" is written into the statute as a named risk. Organizations that can already produce a record of what their agents did, and when a control failed, will be better placed whichever version of federal rules emerges.
Source: Politico: Why Anthropic's draft bill sets $1 billion threshold for AI risk | Politico: Anthropic quietly pitches its own AI safety bill to lawmakers | Reuters via Investing.com