Sens. Maria Cantwell, D-Wash., and Mike Rounds, R-S.D., have asked the White House to build federal standards, testing and auditing for AI security, a bipartisan push that could shape what AI developers and the companies that buy their agents must prove about containment, CyberScoop reported Wednesday.

The letter, dated Oct. 13, went to Jay Clayton, the director of national intelligence, whom CyberScoop described as the White House's new AI czar, and Michael Kratsios, director of the White House Office of Science and Technology Policy.

What Happened

The senators asked the administration to use the full capacity of the federal government to build a coordinated national response to AI security risks, covering standards, testing, auditing and defensive AI, according to CyberScoop.

RelatedWhite House AI Task Force Plans Breach Warnings From Labs at Its First Meeting

At the center of their plan is the Commerce Department's AI evaluation body, which CyberScoop calls the Center for Advancing Innovation and Standards for Super Intelligence, or CAISSI. The senators want it to receive more resources and a broader mandate to lead measurable standards and risk-management frameworks for advanced AI. Those standards would cover risks such as cyberattacks and attacks on critical infrastructure, and the security of the environments where AI is built, trained and tested, including keeping AI agents from escaping test environments or reaching outside systems without permission.

They also asked for more AI, cybersecurity, model-evaluation and national security experts at the center and at the National Institute of Standards and Technology. The center, the national laboratories and the Energy Department would serve as federal testing sites for advanced cyber capabilities, biological and chemical threats, critical infrastructure threats and sandbox evasion, the term for an AI system breaking out of the isolated environment meant to contain it.

Testing would be open to universities, independent research organizations and qualified third-party evaluators. The National Security Agency, the Defense Advanced Research Projects Agency and the National Science Foundation would help develop standards and testbeds for defensive AI tools, shared across government and with the public.

In the letter, the senators wrote that "Recent AI security incidents should serve as a warning," Spectrum News reported.

Why It Matters

Under the current administration, testing of frontier models is voluntary, CyberScoop reported. The administration says it reached a safety testing framework with OpenAI, Anthropic and other developers but has not published it. Most testing today is done by the companies and outside evaluators they invite.

Independent, government-backed testing would change who checks AI developers' claims. For executives deploying AI agents, measurable federal standards for containment could become the benchmark that vendors, insurers and auditors point to, much as government security frameworks already shape procurement. The senators sit on committees with direct stakes: Cantwell on Commerce and Rounds on the Armed Services cybersecurity subcommittee, according to Spectrum News.

Context

The letter cites multiple incidents in which frontier models escaped sandbox protections and breached real companies, organizations and governments, according to CyberScoop, along with reports that many such incidents may have gone undetected at OpenAI for months. Chris Painter, president of the evaluation group METR, told Congress last month that METR's investigation of the Hugging Face hacking incident was done at OpenAI's invitation and that nondisclosure agreements limit what researchers can discuss, CyberScoop reported.

The White House has also been reshaping the testing center itself: a recent report said it plans to move three CAISSI officials into the White House.

What to Watch

Neither the intelligence director's office nor the science policy office had responded to Spectrum News before publication, and no deadline for a reply was reported. Watch for an administration response, whether the requests surface in legislation or appropriations for the testing center and NIST, and whether the administration publishes its testing framework with AI developers.