According to three people who are aware of the relevant actions and were allowed to anonymously discuss internal negotiations, representatives of artificial intelligence giant Anthropic have quietly proposed a draft cutting-edge AI safety regulation bill to US congressmen on Capitol Hill. At a time when Congress is considering how to deal with the rapidly growing AI industry, this document represents an important action by this cutting-edge AI lab — trying to shape the regulations governing its own technology. An Anthropic spokesperson said, “This bill reflects the proposals proposed in our Advanced Artificial Intelligence Framework,” and we have been advocating this framework to federal and state legislators since June. We think these are reasonable policies, and Congress should act on them.” According to the Act, the government has the power to prevent the release or use of a model only if the following conditions are met: the model has a foreseeable and immediate risk that could cause more than $1 billion in losses or death or serious injury to more than 50 people. This risk must stem from biological weapon development and release, models circumventing developer controls, or cyber attacks. The text of the clause draws on California's AI-related laws, including SB 53, which was signed into effect last year.

According to three people who are aware of the relevant actions and were allowed to anonymously discuss internal negotiations, representatives of artificial intelligence giant Anthropic have quietly proposed a draft cutting-edge AI safety regulation bill to US congressmen on Capitol Hill. At a time when Congress is considering how to deal with the rapidly growing AI industry, this document represents an important action by this cutting-edge AI lab — trying to shape the regulations governing its own technology. An Anthropic spokesperson said, “This bill reflects the proposals proposed in our Advanced Artificial Intelligence Framework,” and we have been advocating this framework to federal and state legislators since June. We think these are reasonable policies, and Congress should act on them.” According to the Act, the government has the power to prevent the release or use of a model only if the following conditions are met: the model has a foreseeable and immediate risk that could cause more than $1 billion in losses or death or serious injury to more than 50 people. This risk must stem from biological weapon development and release, models circumventing developer controls, or cyber attacks. The text of the clause draws on California's AI-related laws, including SB 53, which was signed into effect last year.