White House AI adviser Sriram Krishnan's 'no FDA for AI' pledge shifts U.S. policy from a pre market gate to ad hoc federal pull, with Anthropic and OpenAI as the working examples.
When then-White House AI adviser Sriram Krishnan told the Financial Times on July 3, 2026 that "there will not be an FDA for AI" under President Trump, he was not describing a retreat from AI governance. He was describing a preferred instrument. The administration is choosing ad hoc federal leverage over a standing pre-market agency, and the past two months have shown what that choice looks like in practice.
On June 13, a sweeping U.S. order forced Anthropic to shut down access to Mythos, the company's latest model. Two weeks later, on June 26, OpenAI said it had limited the rollout of GPT-5.6 after a government request, then publicly argued such restrictions "shouldn't be the norm." Those two interventions are the working test of the policy Krishnan described. The administration is not leaving frontier models alone. It is choosing which models to touch, when, and through what channel, and declining to write that discretion into a permanent rule.
Wire coverage frames this as a deregulatory headline, but the underlying power to constrain a model before it ships is intact. Tech Policy Press reports the administration has formally abandoned the FDA-style proposal, and Krishnan's rhetoric matches. He called a centralized pre-market gate "sand in the gears" of progress and said it "is never, never going to happen under President Trump," according to the Financial Times interview summarized by PYMNTS. The distinction is between bureaucratic form and underlying constraint.
The policy anchor is Executive Order 14409, signed June 2 under the title "Promoting Advanced Artificial Intelligence Innovation and Security." It describes a deregulatory posture but does not explicitly authorize model-by-model intervention. The leverage for the Mythos and GPT-5.6 actions must therefore run through other channels: procurement access, export controls, or informal pressure on cloud and compute providers. Semafor reported on June 27 that Mythos access was selectively restored to some U.S. companies, a release pattern consistent with case-by-case executive discretion rather than a published rule.
Krishnan's posture has a domestic political backdrop he did not address in the FT interview. Polling cited via the Financial Times shows a majority of Americans support tough AI regulation, a political environment that a standing FDA-style agency would have given a named regulatory target. Keeping the lever informal lets the administration continue to act while claiming the opposite.
Krishnan, a former venture capitalist who previously worked with Elon Musk, has been among the leading internal voices for a laissez-faire line, according to Times of India coverage. His "doomer" critique of AI safety advocates maps onto a worldview in which public regulatory institutions slow progress, a frame summarized by AI Weekly in its read on the FT interview.
The corporate side is building its own de facto limits in parallel. PYMNTS Intelligence's latest report finds companies restricting AI agents to read-only "look-up" access: 100% of wholesale firms surveyed, 90% of retail, 85% of construction. No sub-industry surveyed fully permits autonomous AI action. The private sector is pulling back even as the public sector reaches in, case by case.
The question Krishnan left unresolved is whether the next model shutdown will come with a public rationale. The Mythos pull and the GPT-5.6 pause both came with explanations shaped more by industry pushback than by written criteria. If the administration wants the discretion to keep working, it will need either a published standard or a louder public defense. Krishnan's July 3 interview did neither.