Anthropic Proposes AI Legislation Bill to US Congress

Artificial intelligence company Anthropic has brought forward a new artificial intelligence safety bill proposal prepared for consideration in the U.S. Congress. The bill requires very high thresholds to be crossed before the government can halt the release or use of an artificial intelligence model.

According to the drafted bill, government intervention would only be possible if an artificial intelligence model carries the risk of causing more than $1 billion in damages or resulting in the deaths or serious injuries of more than 50 people. Risks that could justify intervention are limited to the production of biological weapons, the model escaping developer control, or cyberattacks; other harms such as artificial intelligence-assisted fraud are kept outside this scope.

The proposal also consolidates oversight authority within the Department of Commerce, stipulating that no U.S. agency other than this department could suspend model development. Among the provisions included in the bill are that independent third parties registered with the Department of Commerce would periodically inspect laboratories and artificial intelligence models, and that these evaluators could be funded by the laboratories provided their independence is not compromised.