Anthropic Secretly Lobbying Government to Only Shut Down an AI Model If It’s Poised to Kill More Than 50 People

In recent months, AI companies such as OpenAI and Anthropic have been funneling millions of dollars into political action groups, aiming to influence electoral candidates ahead of the US midterm elections. While the exact financial contributions are known, the specific policy stances of these tech giants have remained unclear until now.

According to Politico, representatives from Anthropic have been in private discussions with lawmakers regarding proposed AI regulation. The ideal legislation, as per Anthropic, would empower the federal government with a "kill switch" to deactivate any AI model deemed to be an imminent threat. However, this threshold for intervention is not arbitrary.

The government can only claim a model poses an "imminent threat" if it risks serious injury or death of 50 or more individuals, or if it results in property damage of over $1 billion. Furthermore, the intervention could occur due to three specific categories of risk: the release of a biological weapon, a catastrophic cyberattack, or if the AI model escapes the control of its creators.

The final decision would rest with a third-party monitor under the Department of Commerce, who would have 72 hours to notify Commerce officials of any potential breach. These officials would then forward the evidence to a federal judge, who would ultimately decide whether to allow the AI model to be released to the public. Anthropic's spokesperson confirmed that this bill aligns with the Advanced AI Framework advocated for by the company since June.