Anthropic Proposes AI Safety Bill with Major Damage Thresholds
Anthropic has outlined a proposed AI safety bill that would establish a high threshold for government intervention in AI development. Under the framework, the Commerce Department would become the primary federal agency empowered to halt an AI model, but only in cases where the system poses what the company describes as "catastrophic" risks.
The proposal sets specific trigger conditions: intervention would occur if an AI model could potentially cause 50 or more deaths or result in at least $1 billion in damages. Anthropic argues this approach targets the most severe, low-probability risks rather than more common harms that might emerge from AI systems in everyday use.
The legislation reflects ongoing debates about how to regulate rapidly advancing AI systems. Anthropic, known for its Claude family of AI models, has positioned itself as a company that prioritizes safety research alongside capability development. By proposing a framework with such a high damage threshold, the company appears to be advocating for a targeted regulatory approach focused on preventing catastrophic outcomes rather than imposing broad restrictions on AI development.
The proposal would effectively concentrate authority over AI model halts within a single agency, the Commerce Department, rather than distributing oversight across multiple regulatory bodies. This centralization, according to the proposal's supporters, could streamline emergency responses to genuinely dangerous AI systems while avoiding slower, more fragmented regulatory processes.
The bill has not yet been introduced in Congress, and the proposal is likely to face scrutiny from lawmakers, civil liberties groups, and competing industry interests as discussions around AI governance continue.