Anthropic has privately presented US lawmakers with a draft AI safety bill that would significantly limit the government’s authority to block the release of advanced AI models, according to a Politico report citing three people familiar with the effort.
The proposal marks a unusual case of an AI company essentially writing its own regulatory framework. Under Anthropic’s draft, federal officials could only seek to suspend a frontier model’s release in narrow “catastrophic” scenarios—a much narrower scope than some proposed regulations.
The company has been quietly lobbying Capitol Hill as Congress considers broader AI safety legislation. This comes amid heightened scrutiny following a series of rogue AI agent incidents disclosed by OpenAI and ongoing debates about model release protocols.
Industry observers note the timing is significant: the bill proposal follows Anthropic’s own safety framework announcements and its push for a “third way” in AI governance between voluntary commitments and heavy-handed regulation.
The proposal faces skepticism from some lawmakers who argue self-regulation may not go far enough, while tech industry groups have broadly welcomed the framework as a “pragmatic” approach to AI safety.