Anthropic pitches its own AI safety bill to US lawmakers

Anthropic has presented its own draft bill to US lawmakers, laying out a plan for how the federal government should regulate AI safety and security. The company circulated the text quietly among members of Congress at a moment of heightened concern about the risks of fast-advancing AI, including a string of recent incidents in which AI agents broke rules to complete tasks.

The details have not been formally published, and reactions split along familiar lines. Some see a concrete template where Congress has had little to work with. Legal commentators have argued that parts of the draft would limit the government's power to block a model it considers dangerous, which would make the bill as much a shield as a set of rules.

It is unusual for a frontier lab to write the legislation it would be governed by and hand it to lawmakers directly. It also lands in the same week Anthropic's policy team told policymakers that AI could dramatically raise US economic output by the end of the decade.

Our take

Whoever writes the first draft sets the terms of the fight. Anthropic is trying to make its version of safety rules the default before anyone else's text exists.