Congress Weighs AI Safety Rules as Anthropic Proposes New Bill

US lawmakers are advancing bipartisan efforts to regulate artificial intelligence as concerns about public safety, economic damage, and increasingly powerful AI systems grow. Proposals discussed in Washington include mandatory safety assessments, emergency shutdown mechanisms, and restrictions on releasing models that pose severe risks, while Anthropic has backed a separate legislative proposal addressing potentially catastrophic AI threats, according to CEPA and Bloomberg.

At CEPA's annual Tech & Security Conference, held before the upcoming midterm elections, Republican and Democratic lawmakers argued that safety requirements could coexist with technological progress. Their proposals seek to establish federal standards for advanced AI developers while addressing concerns about inconsistent state regulations and inadequate oversight.

According to Bloomberg's October 16 report, citing Politico, Anthropic proposed legislation that could allow the government to block an AI model's release if authorities determine it poses an imminent risk of killing or injuring more than 50 people or causing more than $1 billion in damages. The reported threshold applies to threats involving the creation or release of a biological weapon, an AI system operating beyond human control, or a cyberattack.

Separately, Representative Nathaniel Moran, a Republican from Texas, is advancing the AI Kill Switch Act with Democratic Representative Ted Lieu. The proposal would require developers of the most powerful frontier AI models to maintain the ability to throttle, suspend, or shut down systems if they threaten to cause $100 million in economic damages or 10 deaths.

Moran compared AI safeguards with vehicle safety requirements, arguing that protective mechanisms can support rather than obstruct technological development.

"The brakes allow you to drive faster, not slower," said Moran. "AI brakes should not slow innovation. They should enable faster movement with a safe backstop."

Representative Lori Trahan, a Democrat from Massachusetts, has partnered with Republican Representative Jay Obernolte on the proposed FRONTIER Act. According to CEPA, the legislation would establish nationwide requirements for catastrophic-risk transparency, incident reporting, and independent auditing, overriding conflicting state-level rules.

Trahan said the legislation would make risk disclosures and independent verification mandatory. It would also establish an Undersecretary for AI Security within the Department of Commerce and create a framework for licensing third-party evaluators. Developers with more than $5 billion in annual revenue would have to use independent evaluators to assess frontier-model risks.

"We need to have catastrophic risk transparency [and] incident reporting that is mandatory and not voluntary, and we need an independent verification model," said Trahan.

Trahan has also proposed the CLAIM Act, which would hold AI companies liable when their models break the law and direct courts to presume that AI models are capable of intent. CEPA noted that comparable liability approaches already apply to products such as drugs and automobiles.

Representative Don Beyer, a Democrat from Virginia, has proposed minimum safety standards for advanced AI laboratories, including a structure modeled on the Federal Reserve that could mandate testing of advanced models. He is also pursuing a master's degree in machine learning at George Mason University to deepen his understanding of the technology.

Beyer acknowledged uncertainty about the scale of potential harm while emphasizing that the underlying risk deserves attention.

"There's this whole spectrum of p-doom = 100 (the probability that AI will cause an existential catastrophe) and 0, and it's probably somewhere in between," said Beyer. "Everyone agrees there's a real risk."

The lawmakers also challenged the Trump administration's less restrictive approach to technology policy. They questioned whether a proposed US-China hotline for AI incidents would provide sufficient safeguards if participation depended primarily on voluntary compliance.

CEPA reported that the lawmakers favored international cooperation, drawing comparisons between potential US-China AI agreements and arms control treaties negotiated by the United States and the Soviet Union during the 1970s and 1980s. They also emphasized cooperation with European allies, whose talent and infrastructure could support shared AI safety efforts.

Another unresolved issue is recursive self-improvement, a process in which AI systems improve their capabilities without direct human prompting. According to CEPA, lawmakers are considering whether such capabilities should be prohibited because a self-improving system could potentially circumvent shutdown mechanisms or manipulate safety tests.

Congress remains divided over the appropriate level of government control, ranging from intervention after a serious incident to requiring regulatory approval before AI models can be released. The competing proposals reflect broader efforts to define enforceable safeguards as advanced AI capabilities expand.

Follow suaragarut.id Add to preferred sources on Google

• AI regulation
• artificial intelligence
• US technology policy
• AI safety