Anthropic Proposes AI Bill With 50-Death Threshold Before Models Can Be Halted
The company has reportedly been working with lawmakers on a plan with few strict rules since early summer.
- By
- Chris Walker,
- Truthout
- Published
Anthropic, one of the leading companies developing artificial intelligence (AI), has proposed a law for Congress to set rules for the technology. The proposal would sharply limit which government agencies could stop AI models and set narrow conditions for when the government could step in.
Politico reported on Friday on details of the proposal, which had not initially been made public.
The proposed law would let the federal government stop an AI model from being released or continuing to operate only if it posed an immediate danger. The proposal says the danger must involve either $1 billion in financial losses or the possibility that an AI model could kill or seriously injure at least 50 people.
The danger must also be directly linked to one of three things: creating a biological weapon, an AI model escaping its developers' control, or some kind of cyberattack.
Independent reviewers working under the Commerce Department would decide whether the conditions for stopping an AI model had been met. AI companies themselves might pay these reviewers' salaries, which could create a conflict of interest.
Only federal judges in the U.S. District Court for the District of Columbia could order action to stop an immediate threat from an AI model. Only the Commerce Department, not any other department or federal agency, could bring claims to the appeals court.
After details of the proposal were leaked, Anthropic released a statement defending it, saying:
This bill reflects the ideas in Anthropic's Advanced AI Framework. Since June, we have been urging federal and state lawmakers to support these ideas. We believe these are sensible policies that Congress should put into action.
However, experts in the industry have said that rules will not truly protect people unless they set stronger standards.
“Without really strong actions that change the incentives companies face when deciding what to test and how to fix problems, I don't think we will deal with the risks that are emerging,” Miranda Bogen of the Center for Democracy and Technology told Mother Jones earlier this summer.
Alex Turner, a former research scientist at Google DeepMind, published an opinion article in The Guardian in September calling for stronger controls.
“We can and should demand that our governments protect us from the disaster of AI that gets out of control. … Today's AI systems sometimes seem to lie or cheat, even when they know better,” Turner said.
Despite strong public support for AI regulation, President Donald Trump has rejected efforts to regulate the industry, calling them a “hoax.” In several posts on Truth Social, Trump has said he does not believe AI poses a threat. This is despite Anthropic acknowledging several possible dangers in the legislation it proposed earlier this year.
“I am the Hoax Buster, and I'm right now breaking another Hoax — That AI is going to take over, consume, and destroy the World, and that Robots will be marching into our Cities, and getting rid of us all,” Trump wrote in a post last month.
In another post in September, Trump also claimed that he himself was the only regulation needed.
“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” Trump wrote.