Anthropic Pitches Draft AI Safety Bill to Lawmakers

The tension between catastrophic risk thresholds and regulatory preemptions

Anthropic proposes a federal standard that allows the government to block AI models only when there is an imminent risk of more than 50 deaths or $1 billion in damages. This framework limits the scope of intervention to specific threats like biological weapons, cyberattacks, or loss of control, excluding other harms such as AI-driven fraud.

The proposal seeks to replace various state laws with a single national standard. This creates a conflict between those who believe a high, catastrophic threshold provides a practical balance for innovation and those who argue it sets the bar for intervention impossibly high. The outcome depends on whether the government accepts a role limited to preventing extreme disasters or insists on a more proactive review of broader societal risks.

This structure mirrors the Glass-Steagall Act, where a clear regulatory line was drawn to separate different types of financial activities to prevent systemic collapse. While that act sought to constrain risk through separation, this proposal seeks to constrain government power by defining the exact scale of a disaster required for intervention.

Between legislative gaps and the risk of industry self-regulation

Coverage diverges on whether this proposal is a sensible blueprint or a corporate shield. Business Insider provides the most detailed look at the mechanics, noting that the bill suggests using independent evaluators who could be paid by the very labs they are auditing. This detail highlights a potential conflict of interest that other outlets overlook.

Mother Jones frames the move as a calculated effort to favor industry self-regulation, while Daily Caller focuses on the political vacuum, arguing that a lack of federal leadership allowed regional laws to set the initial stage. While the outlets agree that the bill aims to preempt state laws, they differ on whether this is a necessary move for efficiency or a strategy to evade stricter local oversight.

Readers should watch whether the proposed thresholds for 'catastrophic risk' are viewed as a floor for safety or a ceiling for government authority. Future reporting needs to examine the specific criteria these independent evaluators would use to determine if a model hits the 50-death or billion-dollar mark before it is released.