Anthropic proposes draft legislation on federal artificial intelligence safety regulations

Artificial intelligence company Anthropic has circulated a draft bill to United States lawmakers proposing new safety guidelines for the industry, TechCrunch and Politico reported. TechCrunch, Futurism, and Cybernews reported that the proposed legislation outlines specific conditions under which government authorities could halt the deployment of an artificial intelligence model. [1][4][5]

According to TechCrunch, Futurism, and Cybernews, the draft proposal includes provisions that would allow regulators to block an artificial intelligence system if it poses a risk of killing more than 50 people. Mezha reported that the draft bill also sets a threshold for restricting a model if it causes at least 1 billion dollars in financial damage. [1][4][5][6]

TechCrunch reported that under the proposed terms, federal agencies would not be allowed to block the release of an artificial intelligence model without approval from a judge. The draft bill is intended as a potential framework for future Congressional action as public and political debate surrounding artificial intelligence technology continues. [1]

Anthropic relies on a technique called Constitutional AI to train its models, including its Claude language model, against natural-language principles. Broader debates surrounding artificial intelligence safety include concerns regarding copyright, misinformation, security, and potential economic disruptions. [7][8]