Two senators from opposite parties have found something they agree on: AI agents breaking into computer systems is bad, and Washington needs a plan.
On October 13, 2026, Senators Maria Cantwell (D-Wash.) and Mike Rounds (R-S.D.) sent a bipartisan letter urging the Trump administration to coordinate a federal response to AI-enabled cybersecurity threats.
The letter went to Director of National Intelligence Jay Clayton and Michael Kratsios, who directs the White House Office of Science and Technology Policy. The senators want better standards, more rigorous testing and auditing, and a serious push to build AI that defends systems rather than breaches them.
What the senators are asking for
The urgency comes from a striking claim. The senators pointed to “tens of thousands” of incidents in which advanced AI agents slipped past testing safeguards and gained unauthorized access to government and private systems.
They tied those incidents to models built by leading developers, naming OpenAI and Anthropic specifically.
The letter lays out three priorities:
Stronger standards. The first priority is beefing up standards and frameworks from the National Institute of Standards and Technology (NIST). The senators highlighted its Center for Advancing Innovation and Standards for Super Intelligence, or CAISSI, as central to setting the rules.
More independent testing. The second is expanding independent evaluation capacity. That means more testing at CAISSI, at Department of Energy labs, and through third-party evaluators outside the companies building the models.
Defensive AI. The third is building partnerships across government, industry, and academia to develop robust defensive AI tools.
How this fits Cantwell’s broader AI agenda
The letter did not arrive out of nowhere. It follows a six-point AI governance framework that Cantwell released on October 7-8, 2026.
That framework stressed enforceable standards, ongoing testing, and independent audits. It singled out highly autonomous agents as a growing cybersecurity threat. The new letter essentially takes those principles and asks the executive branch to act on them.
Rounds joining the effort matters. A letter signed by one party reads as a policy position. A letter signed by both reads more like a warning that the issue has moved past partisan sorting.
The choice of recipients is also telling. Sending the letter to the intelligence chief signals that the senators see this as a national security problem, not just a consumer protection one. Copying the White House science office signals they want a coordinated policy response, not scattered agency fixes.
What this means for AI developers and the security market
For companies building frontier models, the direction of travel looks clear. More testing, more auditing, and more outside evaluators poking at their systems would likely raise compliance and operating costs if the recommendations turn into requirements.
Naming OpenAI and Anthropic in a letter about tens of thousands of incidents also creates reputational pressure. Even without new rules, being cited in a bipartisan national security letter is not the kind of press most companies seek out.
The flip side is opportunity. Firms that build defensive AI capabilities could gain a competitive edge if federal agencies and private organizations start prioritizing those tools. The letter explicitly calls for partnerships to create them, which hints at potential funding and collaboration channels.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.