Anthropic bill stops an AI model only at $1B damage risk
The government can stop a release only if a model risks $1 billion in damage or more than 50 deaths.
Anthropic wants a US bill on AI safety, Politico reports. In the bill, the government can stop the release of an AI model only on one condition. The model must have a risk of $1 billion in damage, or of more than 50 deaths or injuries. An Anthropic spokesperson says the bill follows the Advanced AI Framework of the company. The company has pushed it with lawmakers since June.
How the bill works
A third party on record with the Department of Commerce checks the lab and its AI models. The lab can pay the checkers, but only if this does not change their independence.
The risk must come from one of three causes:
- the creation or release of a bioweapon
- a cyberattack made by AI or with AI
- a lab that cannot control a model or agents
The process
If the checker sees a risk of one of these causes, it must tell the department in 72 hours. The department then gives the signs to a federal judge in the District Court for the District of Columbia. The judge makes the decision on the release.
Only the Commerce Department can tell a lab to stop a release. No other agency or officer in the US can do it.
What is not known
The bill is a proposal. The report gives no date for a vote. It does not name the checkers. Congress also works on a different proposal for kill switches on the most powerful models.
What to do
For leaders who buy AI: only a judge can start a release stop, and the damage limit is high. Look at how your vendors show the risk level of a model.
Sources
Posted
