Anthropic's leaked federal AI bill draws widespread criticism from AI safety advocates for being significantly weaker than the bipartisan FRONTIER Act

State laws attempting to regulate AI will be superseded by this federal bill if enacted.

A leaked draft of a federal AI bill circulated by Anthropic has drawn significant criticism from AI safety advocates, who claim it is weaker than existing bipartisan proposals like the FRONTIER Act. Critics argue the bill, which was published by Politico, could make it legally more difficult to prevent dangerous AI models from being released. Key points of contention include the lack of standards or licensing for third-party AI evaluators, who can self-certify, potentially leading to weaker assessments. The bill also imposes a high burden on the Commerce Department to block a dangerous model, requiring a court petition with detailed justification and facing a legal standard that does not defer to evaluators and must consider the economic impact of blocking. Furthermore, the draft includes broad preemption of state AI laws, hindering states' ability to enact their own regulations on frontier AI development, including stopping model releases on catastrophic risk grounds. While the bill includes some provisions for security protocols and risk management, critics contend these are overshadowed by its perceived weaknesses in regulatory oversight and preventative measures.

Anthropic's leaked AI bill proposes a regulatory framework that critics argue prioritizes business interests over genuine AI safety. The bill's provisions for third-party evaluators lack necessary standards, allowing self-certification which may lead to superficial assessments. This approach, combined with the high legal bar for blocking dangerous models and the preemption of state laws, creates a system where preventing catastrophic AI risks becomes more challenging than under current legal conditions. The bill's rigidity, preventing federal agencies from setting evolving standards for AI safety, further exacerbates concerns about its long-term effectiveness.

The backlash against Anthropic's proposed legislation signals a deep divide within the AI policy landscape, highlighting the tension between commercial imperatives and existential risk mitigation. Advocates fear that this weaker bill could set a precedent, potentially shutting down opportunities for more robust federal regulation and leaving the public vulnerable to rapidly advancing AI capabilities. The situation underscores the influence of corporate lobbying on nascent AI governance, suggesting that companies may be more invested in shaping regulation to their advantage than in proactively addressing profound safety concerns.