Anthropic's Vision for Federal Regulation
Recent reports revealed that Anthropic has submitted a draft of a new bill to lawmakers in the United States, aiming to establish a regulatory framework for artificial intelligence. This move comes at a time of mounting global pressure to control the pace of advanced model development, as the company seeks to define clear scenarios that justify preventing the release of any AI model into the market.
Catastrophic Risk Standards
The proposed bill defines specific cases classified as an "imminent and catastrophic risk," which are situations that could lead to the death of more than 50 people or cause material damage exceeding $1 billion. The draft stipulates that these risks must stem from:
- The creation or launch of biological weapons.
- The model escaping the control of its developers.
- The launching of large-scale cyberattacks.
These standards are largely similar to those contained in California's Senate Bill 53, indicating an attempt to unify safety standards at the federal level.
What Changes Practically?
The most controversial point lies in the implementation mechanism; the law proposes relying on "independent evaluators" registered with the U.S. Department of Commerce to assess model risks. However, the law grants developers the right to pay these evaluators, provided that the fees are not linked to the evaluation results, which raises questions about the independence of this process.
Furthermore, the law imposes strict restrictions on federal agencies; it prevents them from unilaterally deciding to stop the launch of any model. Instead, if an agency detects an imminent risk, it must notify the Department of Commerce within 72 hours, which in turn must seek a court order from a judge in Washington, D.C., to halt the launch. This means removing direct discretionary authority from technical regulatory bodies and limiting it to the judicial path.
The Context of Regulatory Competition
This initiative comes amid a divide within the tech sector; while some companies call for voluntary regulatory frameworks based on internal monitors, other industry leaders, such as Jensen Huang, CEO of Nvidia, believe that the sector does not need new laws that stifle innovation. Anthropic's move is a proactive attempt by major companies to shape the rules they will be subject to, rather than leaving them to lawmakers who might impose stricter or more comprehensive restrictions.