Anthropic has quietly begun pitching Capitol Hill lawmakers on confidential draft legislation aimed at regulating frontier artificial intelligence (AI) safety, marking a major push by a leading industry player to influence federal oversight.
The unreleased proposal, obtained by Politico, mirrors the safety standards outlined in Anthropic’s Advanced AI Framework. An Anthropic spokesperson confirmed the outreach, calling the draft a set of “sensible policies we believe Congress should act on.”
Under the draft bill, federal authorities could block the release or deployment of an AI model only under strict, high-bar conditions. Interventions would require a foreseeable, imminent threat of more than $1 billion in damages, or the risk of killing or seriously injuring over 50 people.
Such risks would need to originate from specific catastrophic scenarios: the creation or release of biological weapons; a cyberattack of massive scale; and an AI model evading its developer’s internal safety controls.
The threshold reflects language from recent state-level legislation, including California’s SB 53. While major AI labs disclosed several incidents over the summer involving AI agents probing or hacking online platforms, none reported to date meet the high threshold of damage required for federal intervention under this proposal.
The legislation outlines a structured enforcement mechanism. Registered independent evaluators, oversight bodies funded by labs without pay tied to evaluation outcomes, would conduct periodic assessments of frontier models. If an auditor detects an “imminent enumerated risk,” they must notify the Department of Commerce within 72 hours.
To halt a risky model, Commerce would need to request an injunction from a federal judge in the U.S. District Court for the District of Columbia. Crucially, the bill explicitly bars any other federal department or agency from unilaterally suspending a model’s development or deployment.
Beyond enforcement thresholds, the bill mandates that AI developers publish transparency reports, protect whistleblowers, and report critical safety incidents. However, it also seeks to preempt certain state-level AI safety laws regarding public disclosures, third-party evaluations, and incident reporting.
State preemption remains a fierce point of contention in Congress.
Tech industry groups advocate strongly for a unified national framework to replace a complex patchwork of state rules. Conversely, progressive lawmakers oppose federal preemption, arguing it threatens robust safety standards already enacted in states like California, Illinois, and New York.
Anthropic’s stealth lobbying coincides with heightened legislative activity surrounding artificial intelligence.
In the House, Reps. Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.) introduced the FRONTIER Act to establish federal authority over risky models. In the Senate, bipartisan negotiations led by Majority Leader John Thune (R-S.D.), Amy Klobuchar (D-Minn.), and Ted Cruz (R-Texas) continue to navigate pushback from lawmakers seeking stronger governmental review powers.
Meanwhile, progressive figures like Sen. Bernie Sanders (I-Vt.) have advocated far stricter measures, including temporary pauses in advanced AI development.
