Inside Trump's AI Turf War: Spy Agencies Are Fighting for Oversight Control

The Trump administration entered office promising to tear up AI regulation and let innovation run free. Nine months later, that philosophy is cracking — and it's turning into an internal war.

A new Washington Post report published today reveals a fierce turf battle inside the White House: U.S. intelligence agencies are pushing to gain significant power over AI regulation, and not everyone in the administration wants to give it to them.

What Changed: One AI Model Upended Everything

The pivot started with Anthropic's Claude Mythos, a frontier model the company unveiled in April. Mythos is extraordinary — and, to national security officials, deeply alarming.

During internal testing, the model demonstrated the ability to autonomously identify and exploit zero-day vulnerabilities across every major operating system and web browser. It found tens of thousands of previously unknown software flaws, most of which remain unpatched. That kind of capability, in the wrong hands, could enable cyberattacks of unprecedented scale and sophistication.

The Trump administration, which had taken a "move fast, don't regulate" approach to AI since January 2025, suddenly found itself staring at something it hadn't anticipated: an AI system that could materially threaten critical infrastructure.

The Split: Commerce vs. Intelligence

The fracture runs deep. On one side: the Commerce Department, which has championed a light-touch, industry-friendly approach to AI governance since Trump took office. On the other: intelligence and national security agencies — the NSA, CIA, and their counterparts — who argue that powerful AI models now pose risks that require their direct involvement.

Intelligence officials want the authority to review "high-risk" frontier AI models before they're publicly released. That means any sufficiently powerful model would have to pass a government evaluation — not just a voluntary one — before deployment.

Commerce officials are reportedly resistant. They worry that mandatory pre-release reviews would slow American AI development, cede competitive ground to China, and represent exactly the kind of regulatory interference the administration promised to eliminate.

The result, per the Post, is a White House that is "sharply split" — with no clear resolution in sight.

Who's Coordinating the Response

National Cyber Director Sean Cairncross has been working to coordinate the government's Mythos response, describing a scramble to get both government and private sector aligned. The administration has described its efforts to ensure the model is "tested left and right" before broader release.

The Center for AI Standards and Innovation (CAISI), operating under the Commerce Department, has already signed evaluation agreements with Google DeepMind, Microsoft, and Elon Musk's xAI — allowing the government to test their models before public release. Those are voluntary partnerships, however, not binding requirements.

The intelligence community wants more. They're pushing for mandatory review processes tied to national security thresholds — and the authority to delay or block releases they deem too risky.

Why This Matters: A Policy Reversal No One Expected

The speed of this shift is remarkable. As recently as six months ago, the administration was rolling back AI safety guidance inherited from the Biden era and signaling hostility toward anything resembling AI oversight. The Register described the pivot bluntly: "Trump jumps from 'anything goes' to 'strict regulation' AI policy."

Fortune reported that the administration is now "suddenly embracing AI oversight ideas it once rejected" — a reversal driven almost entirely by the capabilities demonstrated by Mythos and the realization that cybersecurity risks from frontier AI aren't theoretical.

The practical implication: the government's entire posture toward AI regulation may be shifting in real time, with national security concerns providing the political cover that purely safety-focused arguments never could.

What's Next: An Executive Order Is on the Table

Multiple outlets report the White House is studying a potential executive order on AI security — one that could formalize some version of the pre-release review process for the most powerful models. Whether that means mandatory reviews coordinated by intelligence agencies or a beefed-up version of CAISI's existing voluntary framework remains unclear.

For AI companies, the stakes are enormous. A mandatory pre-release review requirement could add months to deployment timelines and give government officials effective veto power over commercial AI products. Companies like Google, OpenAI, and Anthropic have invested billions on the assumption that the regulatory environment would remain permissive.

For the intelligence community, this is a moment of institutional opportunity — a rare chance to expand influence over a technology that is rapidly reshaping both national security and the economy.

The Bottom Line

The Trump AI policy era was supposed to be simple: get government out of the way. Mythos complicated that story by revealing capabilities that even deregulation-minded officials couldn't ignore. What's playing out now is a policy reckoning — and the outcome will determine who gets to decide what kinds of AI the world is allowed to use.

Watch this space closely. An executive order, a CAISI mandate, or a negotiated framework between Commerce and the intelligence community could land within weeks. The future of frontier AI governance may look very different by summer.