Anthropic has proposed a draft AI safety bill to the U.S.
Congress, outlining a framework to halt model deployment if independent evaluators identify a foreseeable, imminent risk of major catastrophe.
The proposal includes thresholds for harm and a legal mechanism for the Department of Commerce to block releases via court order.
The framework also suggests a model for third-party evaluation, prompting discussions regarding corporate-funded oversight mechanisms.
The risk must involve more than 50 people who could lose their lives or more than $1 billion in property damage. It must also be linked to the creation or release of a biological weapon, a model that bypasses its creator's controls, or an attack on computer systems.
The draft would require independent evaluators, who would be registered with the U.S. Department of Commerce. AI companies would be able to pay them, as long as their compensation is not tied to the outcome of the evaluation. However, as Politico notes, it is unclear how the payment would be ensured that their judgment is not influenced.
If the evaluators find an imminent risk of major disaster, they will have 72 hours to notify the Commerce Department. The department would then have to ask a judge in Washington to stop the model from being released. The draft prohibits a federal agency from stopping it on its own.
READ ALSO
OpenAI acquires Orchestra Bio and will integrate its team into Rosalind
An Anthropic spokesperson confirmed the draft to Politico and said it is based on the company’s Advanced AI Framework proposals. Anthropic says it has been pitching these proposals to federal and state lawmakers since June.
The 50 people and $1 billion thresholds are similar to those in California's Senate Bill 53, which recently took effect.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
