Lawmakers clash on safety testing language for in-development AI legislation, people familiar say

An in-the-works artificial intelligence regulatory bill from members of the Senate Commerce Committee — sponsored by Amy Klobuchar, D-Minn., and John Thune, R-S.D. with input from Chair Ted Cruz, R-Texas — is facing pushback from the panel’s ranking member, Sen. Maria Cantwell, D-Wash., according to multiple people familiar, due to disagreements over the proposal’s AI model safety testing framework. 

Multiple people familiar with the bill’s negotiations told Nextgov/FCW that Cantwell’s longstanding position supporting strong federal AI safety standards clashes with provisions laid out in the measure that is under development.

“[Cantwell’s] team has been working with the goal of strengthening the proposal’s testing regime,” one of the people familiar with the proceedings told Nextgov/FCW. 

A Democratic committee aide said that the elements in the draft bill that Cantwell wants reworked hinge on having safety tests of AI models be conducted by the companies themselves, who would later present results to the Commerce secretary for deployment approval. Cantwell instead supports language that mandates models instead go through federal agencies — such as the national labs and national security agencies — for security vetting and testing. 

“Cantwell really wanted to get these large language models tested and to have mandatory vetting, but the Cruz-Klobuchar-Thune version now that she is opposing, it's primarily a voluntary standard type situation,” the aide told Nextgov/FCW. “Her position all along has been: we need to address catastrophic risk, but we need to do this in a serious way.”

The aide said Cantwell’s office offered language to Klobuchar’s office and described Klobuchar’s office as “very receptive,” but that Cruz’s approval will be a deciding factor. 

Klobuchar said in a statement to Nextgov/FCW that AI implementation with little to no oversight comes with “significant risk” and that Congress must act. 

“I’m continuing to work toward a bipartisan agreement on legislation for government oversight of the greatest risks posed by AI models,” Klobuchar said. “That includes requiring developers to work with government experts to verify and test models to make sure AI is safe. We cannot allow the release of dangerous models, including those that evade the control of their developers.” 

A second person familiar with the matter said Cantwell — along with Anthropic and AI safety groups — are “refusing to play ball,” and that the bill’s provisions already include AI safeguards focusing on catastrophic harm and safety evaluations. 

Anthropic did not return a request for comment by the time of publication.

“Republicans have created legal duties for companies to manage catastrophic risks and ensure that government can verify they are doing so and doing so in a way that actually increases safety,” the second person told Nextgov/FCW regarding the legislation. “They are unwilling to create a bipartisan product to give the federal government the power it needs while ensuring continued innovation and due process in the courts. Senator Cantwell is going as far as to require mandatory pre-deployment testing/fixes before down-dais Democrats can support.”

One AI safety group said that their priorities are Congress championing state efforts to promote guardrails in the face of federal preemption.

“We need pre-deployment testing of frontier models and strong national standards that keep pace with fast-growing capabilities and risks,” the safety group told Nextgov/FCW. 

While Cruz’s bill with Thune and Klobuchar has yet to be released, Cruz spoke on Wednesday about how AI frontier labs, like OpenAI and Anthropic, are developing their tools and said the alleged lack of safety that a former OpenAI and Anthropic employee detailed in an X post Sept. 8 is “highly concerning,” and that the forthcoming legislation with Thune and Klobuchar will address “catastrophic risk” associated with AI. 

“This is scary stuff, but we’re also not going to be able to stick our head in the sand and pretend technology isn’t happening,” Cruz said on Wednesday. “So we’ve got to put some guardrails on it.”

Scores of lawmakers have issued calls for the federal government to effectively regulate the emerging technology, following the former employee’s allegations.

Cruz noted that, despite a need for safety parameters, it is still critical that the U.S. lead the world in AI development, saying that the nation needs to beat China in developing and deploying the latest advanced AI. 

Cantwell took to X on Thursday to reiterate her opposition to a “weak federal standard” that undermines existing state AI legislation. 

“Republicans are suddenly waking up to the urgent need for tough AI safety rules, a sharp reversal from last year when House Republicans and Chairman Cruz tried to impose a 10-year moratorium on state AI regulations that the tech bros wanted. I welcome the urgency,” Cantwell posted. “Meaningful legislation would require the most powerful AI models undergo testing by scientists and experts at our national laboratories to assess whether they could enable sophisticated cyberattacks or aid the development of biological or nuclear weapons.”

Klobuchar and Thune previously collaborated on the Artificial Intelligence Research, Innovation, and Accountability Act of 2023 that called for strong transparency and risk management reporting requirements from companies deploying AI. It did not pass during that session of Congress but has informed the new bill with Cruz’s involvement, though it is likely to be “a fundamentally different bill,” the Democratic committee aide said. 

A markup scheduled before the August recess that would’ve overseen a version of the Thune-Klobuchar bill was cancelled, with Cruz saying that he wanted to focus on child safety legislation. 

Cantwell’s stance comes as the public discourse on AI has centered on allegations from a former employee of both OpenAI and Anthropic who said that neither company is acting responsibly while developing advanced AI. In response, scores of lawmakers issued calls for the federal government to effectively regulate the emerging technology.