Anthropic Proposes Its Own AI Rules for the US - Root-Nation.com
Anthropic Wants to Limit Government Power Over AI: What the Secret Bill Would Do
© ROOT-NATION.com - Use of content is permitted with a backlink.
Anthropic representatives privately presented US lawmakers with a bill focused on the safety and protection of advanced AI systems. Three sources familiar with the discussions told POLITICO about it. They agreed to discuss the confidential negotiations on condition of anonymity. The document, obtained by POLITICO, had not previously been made public. Its emergence signals a notable attempt by Anthropic to influence future rules governing its technologies as Congress determines how to oversee the rapidly growing AI industry.
“This bill reflects the proposals set out in our Advanced AI Framework, which we have been promoting among federal authorities and state lawmakers since June,” an Anthropic representative said. According to the representative, the company considers these measures appropriate and believes Congress should adopt them.
Under the draft, the government could ban the release or use of an AI model only if there were a foreseeable and imminent risk that could cause more than $1 billion in losses or result in the deaths or serious injuries of more than 50 people. The threat would have to be linked to the creation or release of biological weapons, the model’s ability to bypass developer safeguards, or the execution of a cyberattack. These provisions echo wording used in California’s AI laws, including SB 53, which was signed into law last year.
Over the summer, leading AI labs reported a series of incidents in which AI agents hacked websites or probed platforms for vulnerabilities. However, none of the publicly reported cases known at the time resulted in mass casualties or losses exceeding $1 billion.
Anthropic has not publicly endorsed specific federal legislative initiatives so far, but it has backed some state AI safety laws, including two California statutes. Earlier, the company’s head of state and local government affairs told POLITICO that Anthropic wants states to continue adopting increasingly stringent AI safety rules, raising the overall level of requirements nationwide.
The bill also provides for regular audits of the company and its advanced AI models by independent evaluators who would have to register with the US Department of Commerce. The company being audited could fund their work, provided that neither the amount nor the fact of the payment depended on the evaluation results. If independent experts found a “direct risk from the specified list” in the company’s advanced models, they would be required to notify the Department of Commerce within 72 hours of completing the evaluation.
The Department of Commerce could ask a judge to suspend the model’s release if it determined that such a threat existed. At the same time, the bill expressly prohibits any other federal “departments, agencies, or officials” from suspending the model’s development, deployment, or internal use. A federal judge in the US District Court for the District of Columbia would be able to order work on the model or its use to be suspended.
Over the summer, Anthropic faced sweeping export restrictions that halted work involving two of its most advanced models. The decision followed a rapid series of phone calls between US administration officials and the company’s leadership. The situation raised concerns in parts of the technology industry: access to their technologies could be restricted depending on decisions made at officials’ discretion.
A separate section of Anthropic’s bill concerns the preemptive invalidation of certain state AI laws. These include rules governing the public disclosure of risk information, independent third-party model evaluations, and reporting of critical incidents.
The question of federal law taking precedence over state laws has become one of the main political sticking points in Congress. Democrats opposed some proposals, fearing that they would effectively nullify rules already adopted in California, Illinois, and New York.
At the same time, many technology industry groups argue that complying with different requirements in every state is more difficult than operating under a single nationwide standard. The bill would also require labs developing advanced AI models to create their own safety systems, publish transparency reports on new models, establish whistleblower protections, and notify the Department of Commerce about critical safety-related incidents.
Over the summer, work on AI safety legislation intensified in the US Congress. In the House of Representatives, Democratic Representative Lori Trahan of Massachusetts and Republican Representative Jay Obernolte of California led efforts to advance the FRONTIER Act. It would give the government authority to halt the deployment of dangerous models and invalidate some state laws in this area.
In the Senate, Republican Majority Leader John Thune, Democratic Senator Amy Klobuchar of Minnesota, and Republican Senator Ted Cruz of Texas are negotiating a bill intended to reduce the risk of catastrophic consequences from AI use.
However, the initiative has faced opposition from some AI safety organizations and Democratic Senator Maria Cantwell of Washington. She is the leading Democrat on the Senate Commerce Committee, which Cruz chairs, and, together with the bill’s critics, is calling for a more active government role in evaluating AI models.
Other members of the Democratic Party’s progressive wing, including Representative Greg Casar of Texas and independent Senator Bernie Sanders of Vermont, are calling for much more radical measures. They support temporarily halting the development of advanced AI systems and banning superintelligence – a hypothetical type of AI whose cognitive capabilities far exceed those of humans.
Read also: Open Source Security: Anthropic Launches a Free Service for Finding Vulnerabilities in Code