A confidential draft of a proposed AI safety bill from Anthropic has been obtained by Politico, revealing a framework that would restrict government intervention to scenarios involving extreme financial or physical harm. The document, which the company has pitched to US lawmakers, outlines specific thresholds that must be met before federal authorities can act against an artificial intelligence model.
High Thresholds for Government Intervention
According to the leaked text, the bill would allow the government to crack down on a given AI model only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. The proposal specifies that these risks must stem from the creation or release of a biological weapon, the model evading its developer’s controls, or a cyberattack.
This approach mirrors elements found in California’s recent AI legislation, suggesting a coordinated effort to shape regulatory standards at both the state and federal levels. A spokesperson for Anthropic confirmed the document’s authenticity, stating that the bill represents the proposals outlined in the company’s Advanced AI Framework. The company has been advocating for these specific policies with federal and state lawmakers since June.
Criticism of Self-Regulation
The proposal has drawn sharp criticism from observers who argue that such frameworks allow tech companies to manufacture consent for self-regulation while avoiding accountability for existing harms. Critics suggest that AI magnates leverage warnings of theoretical, wide-reaching catastrophe to expand their influence while doing little to address the immediate negative impacts of their technology on industries and human psychology.
Concerns about the adequacy of current safety measures were highlighted earlier this year when Anthropic’s own safety lead resigned, leaving a note stating that “the world is in peril.” Additionally, OpenAI recently faced accusations from eight US state attorney generals for ignoring its own safety warnings in favor of profit. Regulations like the one Anthropic has proposed arguably give tech companies the appearance of capitulation while only restricting them on their own terms.
The leaked document underscores the ongoing tension between AI developers and regulators. While Anthropic frames its proposal as a sensible policy that Congress should act on, opponents view it as a mechanism to limit government oversight to only the most catastrophic, and therefore rare, scenarios. This leaves a wide gap in regulatory coverage for the myriad of other risks associated with advanced AI systems.
Source: PC Gamer
