Anthropic secretly lobbies Congress to advance legislation on frontier AI regulation
① At a time when AI development is mired in controversy, Anthropic is seeking to have its position incorporated into AI regulatory frameworks. ② According to the draft, the government may only halt the release or use of a model if it could potentially cause losses exceeding US$1 billion or result in the death or serious injury of 50 or more people.
Following its warnings about the risks of AI going out of control and its call to pause cutting-edge AI research, Anthropic has taken another step.
According to the latest reports, representatives from Anthropic recently privately briefed U.S. lawmakers on a draft bill aimed at regulating cutting-edge AI safety. This suggests that, amid growing controversy over AI development, the AI giant is taking a series of steps to influence the formulation of regulations governing its own technologies.
According to the draft bill, the government may prohibit the release or use of a model only if it poses a "foreseeable and imminent risk" that could result in losses exceeding $1 billion or cause the death or serious injury of 50 or more people. Such risks must stem from the development or deployment of biological weapons, the model's circumvention of developer‑imposed controls, or the execution of cyberattacks. The language used in the bill is consistent with California's AI regulations, including Senate Bill 53, which was signed into law last year.
It is worth noting that, despite a series of AI-related safety incidents recently disclosed by leading companies such as OpenAI and Anthropic, none of the cases reported to date have resulted in large-scale fatalities or losses exceeding US$1 billion.
Furthermore, according to the draft, an independent assessment body would be required to conduct periodic evaluations of AI laboratories and their cutting-edge models. Such an assessment body must be registered with the U.S. Department of Commerce and may be funded by the relevant laboratory; however, its remuneration must not be contingent on the evaluation results.
If an assessment agency determines that a cutting-edge model developed by a company poses an "imminent risk" as defined in the legislation, it must report this to the U.S. Department of Commerce within 72 hours of completing the assessment. Upon finding that a particular model constitutes a risk, the Department of Commerce may petition a judge to prohibit its release. However, the legislation expressly prohibits any other U.S. "department, agency, or official" from halting the development, deployment, or internal use of such a model. A judge of the U.S. District Court for the District of Columbia, on the other hand, may issue an order to suspend the model.
"Evaluation bodies" are something Anthropic has long sought to implement. Earlier, when Anthropic CEO Dario Amodei called on the entire industry to slow down the development of cutting-edge AI, one of the three-step proposals he put forward included deploying "embedded evaluators."
Moreover, the company has already taken concrete steps. This Friday, reports indicated that Anthropic has granted an external evaluation firm "employee-level access" to assess the risks associated with its models. The authorized evaluator, METR—a nonprofit organization that has previously collaborated with leading labs such as OpenAI, Anthropic, Google, and Meta—has now deployed a team to the Claude model development team's office and obtained access to Anthropic's hardware and internal systems.
The bill proposed by Anthropic would also carve out certain state-level AI laws from its scope—laws that mandate the public disclosure of risk information, third-party assessments, and reporting of significant safety incidents—and require cutting-edge AI labs to develop safety frameworks, issue transparency reports for newly released frontier models, establish whistleblower protection mechanisms, and report major safety incidents to the Department of Commerce.
Editor/rice