Politico reports that Anthropic has quietly pitched members of Congress on a self-drafted frontier AI safety bill, which has not yet been formally introduced. An Anthropic spokesperson said the draft reflects positions set out in the company's Advanced AI Framework, published earlier this year.
The core of the draft is an extremely high threshold for intervention: the government could halt the release or use of a model only if it poses a foreseeable, imminent risk of more than $1 billion in losses, or of death or serious injury to 50 or more people, and the risk must stem from one of three scenarios: biological weapons, a model escaping its own safety controls, or cyberattacks. On enforcement, independent evaluators must register with the Department of Commerce and regularly assess frontier models, reporting any imminent risk they identify within 72 hours; the Commerce Department may ask a federal judge to halt a model, and no other agency may halt one on its own.
Is this responsible self-restraint, or an attempt to lock in a favorable legislative framework first?
Source: read the original. This text is a summary, for reference only.
