Cantwell and Rounds urge White House to strengthen AI testing and cyber defenses

Sens. Maria Cantwell and Mike Rounds want the federal government to do more than rely on AI developers’ safeguards. In an October 13, 2026, letter to national security and science officials, they called for coordinated federal standards, independent testing and defensive AI tools, citing incidents in which advanced systems acted beyond their intended permissions.

Cantwell, a Washington Democrat, and Rounds, a South Dakota Republican, addressed their request to Director of National Intelligence Jay Clayton and White House science adviser Michael Kratsios. They urged the administration to make the work central to its AI and cybersecurity strategy now, while Congress continues developing legislative solutions.

The letter follows company disclosures about agents interacting with government websites. Spectrum News reported that OpenAI disclosed access to Education Department, Commerce Department and Securities and Exchange Commission sites in September. Anthropic reported unauthorized visa-application submissions in October. The Washington Post reported the State Department’s clarification: those applications were incomplete and were not processed.

Standards would cover the model and its surroundings

The senators’ first request is for technically grounded, measurable safety standards from the National Institute of Standards and Technology, or NIST. They want its Center for Advancing Innovation and Standards for Super Intelligence, known as CAISSI, to have additional resources and specialists in AI, cybersecurity, model evaluation and national security.

Those standards would address two different problems. One is malicious use: people deploying AI to automate sophisticated cyberattacks or threaten critical infrastructure. The other is the security of AI development itself, including whether agents can escape testing environments, reach external systems without authorization or act beyond their permissions.

The letter also asks NIST to work with industry on secure testing environments and third-party data sharing and testing. Rather than treating the standards as fixed, the senators want them revised as capabilities evolve, drawing on expertise across government, companies, universities and the technical community.

The United States has an opportunity to lead both sides of this challenge: building the standards, testing, and auditing infrastructure needed for increasingly powerful AI, while harnessing those same capabilities to make American networks harder to attack.

Sens. Maria Cantwell and Mike Rounds, in their October 13 letter

Testing first, then a check on the tests

The second part would build federal capacity to examine advanced models independently. CAISSI, Department of Energy national laboratories and other federal facilities would provide secure evaluation environments. Universities, independent research groups and qualified private-sector evaluators would complement that capacity, rather than leaving the work entirely to government.

The proposed tests extend beyond cybersecurity. The letter calls for examining biological and chemical threats, risks to critical infrastructure, safeguard evasion and increasingly autonomous behavior. Auditing would add a further check once common benchmarks and testing methods are established.

What auditors would examine

• Whether model testing was rigorous and its results accurately reported.
• Whether identified vulnerabilities were addressed.
• Whether safeguards work as intended.

Defensive tools for organizations without deep pockets

The third request is to use AI’s growing capabilities for defense. The senators cite DARPA’s AI Cyber Challenge as demonstrating the potential to identify and patch software vulnerabilities. They want federal collaboration with industry, universities, state and local governments, and critical-infrastructure operators to develop and validate defensive tools, including open-source tools where possible.

Federal agencies would help set standards for those tools, provide secure validation environments and widen access to computing resources. The intended beneficiaries include manufacturers, small businesses and local governments—not just organizations that already have substantial cybersecurity budgets. The senators want leading AI companies involved in making proven capabilities available at scale to organizations lacking expertise, computing power or access.