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

For months, leaders at major AI companies like OpenAI and Anthropic have spent tens of millions of dollars on political action groups. They are clearly trying to win support from political candidates before the US midterm elections. Read more about their efforts to gain political support.

We have a good idea of how much money major AI companies have spent so far: at least $55.7 million. But it has been harder to know exactly what rules they want the government to make. That may have changed now.

According to Politico, a team from Anthropic has been having private talks with lawmakers about the wording of a draft bill to regulate AI. Three people who know about these talks told the news outlet that Anthropic wants the federal government to have the power to shut down any AI model that poses an immediate threat.

But there is an important limit. For the government to say that an AI model poses an “imminent threat,” it would have to risk seriously injuring or killing at least 50 people, or causing more than $1 billion in property damage. Only three kinds of danger would allow the government to step in: the release of a biological weapon, a massive cyberattack, or an AI model escaping the control of the lab that created it.

Politico explains that an outside monitor would make the first assessment. This monitor would work under the Department of Commerce. If the independent watchdog found that an AI model about to be released was likely to meet these danger conditions, Anthropic's proposal would give the monitor 72 hours to report the concern to Commerce officials. Those officials would then decide whether to take the evidence to a federal judge.

After all these steps, one judge would decide whether the AI model could be released to the public. This process involves many conditions and several government officials and outside groups. It appears to give AI companies a lot of freedom to make decisions for themselves, which seems to be exactly what they want.

“This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June,” an Anthropic spokesperson told Politico. “These are sensible policies we believe Congress should act on.”