Anthropic drafts legislative proposal targeting narrow federal AI safety thresholds

Tech Insider, Bloomberg, and CNA report that artificial intelligence company Anthropic has quietly submitted a draft legislative proposal to United States lawmakers. Politico first reported on October 16, 2026, that the proposal would establish federal guidelines for regulating frontier artificial intelligence models. The draft proposal had not been formally introduced in Congress at the time of reporting. [1][8][9]

Under the terms outlined in the draft legislation, federal regulators could temporarily halt the release or deployment of a model only under specific catastrophic circumstances, according to Tech Insider and LatestLY. Tech Insider reports that these conditions require an estimated damage threshold of $1 billion or a minimum of 50 human deaths linked to the model. [1][6]

LessWrong AI reports that the leaked draft shares features with the proposed FRONTIER Act. The publication notes that Anthropic's proposal incorporates stronger preemption of individual state artificial intelligence laws and expanded transparency mandates, while featuring weaker enforcement mechanisms relative to other legislative proposals. [4]

As FTMQ SI reported earlier, Anthropic previously voiced support for statutory oversight frameworks, having advocated for binding rules during British parliamentary hearings earlier in October 2026. Tech Insider reports that the current draft bill was presented privately to Capitol Hill staff rather than through a public policy campaign. [1][16]