New Bill Proposal for Artificial Intelligence Safety from Anthropic
Anthropic has prepared a legislative bill that could halt artificial intelligence models in cases involving biological weapons, cyberattacks, or loss of control risks resulting in more than 50 deaths or $1 billion in damages.
Amid ongoing safety debates in the field of artificial intelligence, Anthropic has drafted a new AI safety bill for Congress to evaluate. The bill envisions preventing the deployment of AI models under certain catastrophe scenarios.
According to the bill, situations such as the production of biological weapons, the model escaping developer controls, or cyberattacks resulting in the deaths of more than 50 people or over $1 billion in property damage are defined as a "foreseeable and imminent risk." It is noted that these criteria share similarities with California Senate Bill 53, which was enacted in the previous period.
In the proposed system, model risk analysis will be conducted by independent evaluators registered with the U.S. Department of Commerce. If the evaluators detect an imminent catastrophe risk, they will contact the Department of Commerce within 72 hours, and the department will be able to request that a Washington D.C. court halt the model. The bill prohibits federal agencies from having the sole authority to halt deployment.
Sources
- techcrunch.com · Link