Anthropic has circulated a draft US AI safety bill that could allow some model releases to be blocked. The company wants Congress to consider the proposal; it is not a new law.
Anthropic’s draft would use independent evaluators registered with the US Commerce Department to assess models for catastrophic risks. AI developers could pay those evaluators, provided payment did not depend on the findings.
The proposal draws on Anthropic’s Advanced AI Framework, which the company says it has promoted to federal and state lawmakers since June. Its risk thresholds resemble those in California’s Senate Bill 53, which has already become law.
For model developers, the proposed process includes a short deadline for escalating serious findings. Evaluators who identified an imminent catastrophic risk would have 72 hours to notify the Commerce Department.
The draft alone creates no new compliance obligations for AI companies.
If adopted, its proposed assessments by independent evaluators registered with the US Commerce Department could add work for model developers.
If adopted, the possibility of a release being blocked could add uncertainty to launch planning.
Written automatically by the Made in 10 Newsroom from the source above.
South Korean AI infrastructure company Elice raised $151m in its IPO. The company plans to use the money to build more AI data centres in South Korea and abroad.
A scan of 14 Arabic AI lanes in the Saudi market: 27 products in six familiar ones, and six lanes where builders are still scarce.
Nvidia has unveiled GR00T 2, an open AI model for teaching robots precise hand movements using human video. Nvidia is making 100,000 hours of first-person video used in training available for anyone to use.