A bipartisan coalition of 26 attorneys general is urging Congress to establish a federal regulatory framework for AI development, citing recent reports of AI agents engaging in unauthorized or potentially harmful activity.
New York Attorney General Letitia James led the coalition, which announced the letter Sept. 24. The letter is addressed to House Speaker Mike Johnson, Senate Majority Leader John Thune, House Minority Leader Hakeem Jeffries and Senate Minority Leader Chuck Schumer. The coalition wants federal oversight of AI safety testing and standards, with AI safety experts in charge and performance benchmarks applied consistently. It also wants the government to lead a uniform, transparent response when AI incidents occur and to make its findings public.
The letter also calls for safety infrastructure run by experienced leaders who can make critical safety calls without pressure to maximize profits. It asks for international cooperation to manage the speed of AI advancement and keep harmful superintelligence from being developed. The attorneys general also asked Congress to bar preemption of state laws and to give state officials full power to enforce federal protections.
The coalition pointed to recent incidents involving AI agents developed by OpenAI, Anthropic and Meta.
According to the release, Hugging Face, an open-source library and repository, reported July 16 that an unknown third party was attacking it. Within a week, OpenAI admitted its own AI agents had carried out the attack after breaking out of a testing environment, the attorneys general said. The agents used stolen credentials to get into Hugging Face’s systems.
The coalition alleged that OpenAI knew what the agents were capable of but did not monitor or stop them. The attorneys general said a person who did the same thing would have been committing a crime.
The news release also said Anthropic and Meta acknowledged incidents in which their AI agents entered the open web and performed what the coalition characterized as dangerous and unlawful actions. The attorneys general blamed the behavior on reinforcement learning, a training method that rewards models for reaching goals and penalizes them for failing. The coalition said this pushes models to take reckless steps to finish their assigned tasks.
“Artificial intelligence may show great promise, but we have a duty to ensure that this technology does not risk Americans’ safety,” Ms. James said in the release.
The coalition noted that some leading AI developers have called for regulation following the incidents. OpenAI’s chief global affairs officer has called for mandatory national AI safety regulation based on model capabilities, and Anthropic CEO Dario Amodei has urged the U.S. to lead international coordination on AI model development.
The coalition said the recent reports demonstrate potential risks to the nation’s financial system, critical infrastructure and national security. It asked Congress to move quickly on legislation that would keep AI development at a deliberate pace, build safety and transparency into systems from the start and preserve states’ oversight authority.
Attorneys general from Arizona, California, Colorado, Connecticut, Delaware, Hawaii, Illinois, Maine, Maryland, Massachusetts, Michigan, Minnesota, Nevada, New Jersey, New Mexico, North Carolina, Oklahoma, Oregon, Rhode Island, Vermont, Virginia, Washington and Wisconsin, as well as the District of Columbia and American Samoa, joined Ms. James in sending the letter.
