Anthropic drafts AI security bill and recommends it to US Congress members

Anthropic has independently drafted a cutting-edge AI security regulatory bill and recommended it to members of the US Congress. The bill sets a risk threshold for government intervention in the release of AI models: when there is a foreseeable immediate risk that AI models may cause over $1 billion in property damage, or result in over 50 deaths or serious injuries, and the risk arises from the manufacture and dissemination of biological weapons, breaking free from developer control, or launching cyber attacks, the government can prevent their release or use. The draft requires the assessment agency to report to the Ministry of Commerce within 72 hours after identifying the risk. After confirmation by the Ministry of Commerce, an injunction can be applied to the court. At the same time, developers are required to test the model and publicly disclose the results, and undergo regular review by independent assessment agencies. (Source: Politico)