Anthropic is lobbying the US Congress to introduce legislation that would grant authorities the power to suspend AI systems posing significant risks. The proposed bill targets scenarios involving potential damage exceeding $1 billion or incidents resulting in death or severe injury to more than 50 individuals.
The regulatory framework focuses on threats related to the development of biological weapons, cyberattacks, or instances where an AI model escapes developer control. Independent experts registered with the US Department of Commerce would be responsible for assessing these risks and recommending legal action to halt specific projects.
Under the proposed rules, developers would be required to establish internal safety divisions and submit regular reports on their AI models. Companies must also notify regulators immediately regarding any major system failures. The bill includes whistleblower protections for employees reporting violations.
The initiative draws inspiration from California's 2025 AI regulations. Anthropic, the creator of the Claude language model family, advocates for these measures as a necessary approach to overseeing the most powerful artificial intelligence systems currently in development.
