Model Risk Lands on AI Firms as Trump Rejects FDA for AI

If there will be no FDA for AI, who carries the risk when a powerful model ships too fast?

That is the real question inside Sriram Krishnan’s blunt message. The White House’s outgoing artificial intelligence adviser told the Financial Times that the Trump administration will not create a federal licensing regime for AI models, according to PYMNTS.

“There will not be an FDA for AI,” Sriram Krishnan said.

The statement does more than reject a new agency. It says the administration sees premarket AI approval as a brake on deployment, competition, and U.S. advantage. Krishnan framed the alternative as speed with selective intervention, not formal permission slips.

“This administration, [the] president, from day one has been against burdensome, onerous, bureaucratic red tape. We are not in the business of picking winners and losers.”

That posture now sits beside a harder fact: the federal government has already intervened in model launches. PYMNTS notes that Krishnan’s comments came weeks after the government forced Anthropic to pull the latest version of its Mythos model and paused OpenAI’s 5.6 launch.

So the White House is not saying AI should face no checks. It is saying those checks should not look like formal licensing.

If there is no FDA for AI, what replaces the gatekeeper?

Krishnan’s answer is clear: not a centralized agency that requires companies to hire lawyers before releasing a model.

He said such a system would put “sand in the gears” of AI progress and added:

“That is never, never going to happen under President Trump.”

The phrase FDA for AI works politically because it gives people a familiar image: a powerful federal body deciding what can reach the market. But AI models don’t behave like static products. They are updated, integrated, fine-tuned, wrapped into other tools, and deployed across different workflows.

That makes a single launch approval less clean than it sounds.

XOOMAR analysis: The administration’s position appears to be that model oversight should happen through narrower channels: voluntary review, national security checks, cybersecurity testing, industry feedback, and direct intervention when officials see a specific risk. That reading is supported by the source material, which describes Krishnan as backing a more laissez-faire approach while also citing recent government action against specific model releases.

The tension is obvious. A no-license regime can move faster. It can also leave more judgment calls to companies until the government decides a case is serious enough to interrupt.

How does the Anthropic and OpenAI episode complicate the no-license message?

The Anthropic Mythos and OpenAI 5.6 actions show that “no FDA for AI” does not mean Washington intends to stay passive.

PYMNTS describes the federal move as an “unprecedented intervention,” citing the FT’s account that the government forced Anthropic to pull the latest Mythos model while pausing OpenAI’s 5.6 launch. Related reporting says Krishnan defended the Mythos halt as taken “very, very reluctantly.”

That distinction matters.

The policy signal is not “ship anything.” It is “don’t make launch permission the default.”

That is a meaningful difference for AI labs, enterprise buyers, and investors. It also leaves open the question of who defines the threshold for intervention.

Why are data centers now part of the AI regulation fight?

Public backlash is becoming harder to separate from AI policy.

PYMNTS cites findings from Data Center Watch showing that a majority of Americans support tough AI regulations, while at least 75 data center projects were halted by local opposition in the first three months of the year. Related reporting says U.S. data center projects worth about $130 billion faced local opposition in the first three months of 2026.

Those figures should not be collapsed into one claim. PYMNTS says 75 projects were halted. The related report says projects worth about $130 billion faced opposition. Both point in the same direction, but they measure different things.

Krishnan reportedly argued that the backlash stems partly from how AI companies explain themselves. Related coverage says he blamed the industry’s messaging, including an emphasis on “dystopian narrative and scenarios,” rather than government policy alone.

XOOMAR analysis: That is the weak spot in the no-license stance. If Washington rejects a visible federal gatekeeper while local communities are already blocking infrastructure, the fight doesn’t disappear. It moves downward, toward zoning fights, state rules, local opposition, and company-level trust.

For readers following XOOMAR’s broader AI coverage, that public trust problem also shows up in experiments like 277 Americans Put AI Collective Intelligence on Trial. It is a different issue from model licensing, but it reflects the same pressure point: people want proof that AI systems can be governed before they are asked to accept their consequences.

Are companies actually trusting AI enough to let it act?

Not fully.

PYMNTS Intelligence research cited in the source says companies are using AI across more parts of their operations, but are limiting what AI agents can do on their own. A wide majority restrict agents to “look-up access,” meaning the system can retrieve information but cannot act on it.

The figures are striking:

• Wholesale firms: 100% surveyed limit AI agents to look-up access.
• Retailers: 90% do the same.
• Construction companies: 85% do the same.
• Autonomous action: Not a single company in any sub-industry fully permits it.

PYMNTS summarized the pattern this way:

“The restriction is not a sign that AI deployment has stalled. The picture is of companies that have embedded AI broadly for research, analysis and document generation, and drawn a line before execution.”

That line matters more than the rhetoric.

The White House may reject an FDA for AI, but enterprise customers are already building their own friction into deployment. They are accepting AI for research, analysis, and document generation. They are not yet handing it full control.

Who benefits from a no-license AI policy, and who absorbs the ambiguity?

AI developers get the clearest near-term benefit. Without formal licensing, model releases face fewer federal procedural barriers.

But ambiguity does not vanish. It shifts.

Founders can move faster, but they still face customer reviews, security questions, and the possibility of direct federal intervention if a model triggers a government concern. Large AI labs may prefer the flexibility, though the Anthropic and OpenAI cases show that scale can attract scrutiny. Enterprise buyers get more tools faster, but must decide where to draw the execution line.

The administration’s posture also leaves state-level pressure unresolved. Related reporting says the administration enacted an executive order in December 2025 preempting conflicting state AI laws, and quoted Krishnan saying:

“We don’t want California to set the rules for AI across the country.”

That is the other front. If the federal government refuses licensing and tries to block a state patchwork, then voluntary frameworks and targeted federal action carry more weight.

This fits a broader pattern in tech policy: Washington can reject one regulatory model while still fighting over narrower rules. XOOMAR has tracked that dynamic in adjacent digital-policy fights, including Senate Threatens to Sink House Kids Online Safety Bill. The details differ, but the governing problem rhymes: national technology markets keep colliding with fragmented political demands.

Which evidence would prove the no-FDA bet is working?

The Trump administration’s bet is that the U.S. can govern AI without a single licensing agency. That case gets stronger if voluntary reviews catch serious issues early, if targeted interventions remain rare and well-explained, and if companies keep limiting autonomous AI action where the risks are not understood.

It weakens if model pauses become frequent, opaque, or politically selective. It also weakens if local opposition keeps halting AI infrastructure while Washington insists formal regulation is unnecessary.

The next fight will not be over whether an FDA for AI exists. Krishnan has made the administration’s answer plain.

The fight will be over the substitutes: voluntary model access, national security review, industry clearinghouses, state preemption, local data center resistance, and enterprise-level controls. If those mechanisms produce confidence, the no-license approach holds. If they don’t, pressure for a harder federal regime will build from the outside in.

Impact Analysis

• The administration is rejecting a formal AI licensing system while still leaving room for federal intervention.
• AI companies may face fewer premarket barriers but continued uncertainty around launch reviews.
• The policy direction prioritizes speed and U.S. competitiveness over centralized approval.