The Trump administration’s departing AI adviser says there will be no licensing regime for artificial intelligence — then explains why the government just killed Anthropic’s most advanced model and stalled GPT-5.6.
The Paradox — By The Numbers
$130B
Data-center investment disrupted by local opposition, Q1 2026 alone
75+
Data-center projects disrupted by community opposition in Q1 2026
2
Frontier models directly intervened on: Anthropic’s Mythos + OpenAI GPT-5.6
0
Published rules governing when the government can pull a frontier model
In his first in-depth interview since departing the administration, Sriram Krishnan — the former White House AI adviser and key architect of the Trump team’s light-touch posture — told the Financial Times bluntly: “There will not be an FDA for AI.” Krishnan, an Indian-born former venture capitalist who worked alongside Elon Musk before entering government and served beside AI tsar David Sacks, framed the stance as ideologically principled: “We are not in the business of picking winners and losers.” No statutory licensing regime. No mandatory pre-market approval. Washington, by his account, would govern AI by getting out of the way.
The problem is what happened in the weeks before that interview. On national-security grounds, the administration intervened to force Anthropic to withdraw Mythos — its most advanced frontier model — and separately held up the release of OpenAI’s GPT-5.6. No published criteria. No formal process. No statute invoked. Two of the most consequential AI deployment decisions of 2026 were made via discretionary executive action, not law. Krishnan has since spent weeks in diplomatic damage control, reassuring world leaders rattled by Washington’s willingness to unilaterally switch off access to Mythos.
The broader picture Krishnan sketched to the FT is one of political friction the administration did not fully anticipate. A large majority of Americans support stringent AI regulation — a public appetite the White House has not neutralised. At least 75 data-center projects worth roughly $130 billion were disrupted by local opposition in just the first three months of 2026, per research body Data Center Watch. Krishnan’s diagnosis was pointed: labs have done a “terrible job” communicating benefits and have “focused so much on the dystopian narrative.” In his telling, the industry’s own doomer messaging created the regulatory pressure Washington is now trying to hold back.
Key Timeline — 2026
Q1 2026
75+ data-center projects worth ~$130B disrupted by local opposition across the US, per Data Center Watch.
Early 2026 — National Security Intervention
Trump administration forces Anthropic to withdraw Mythos on national-security grounds; OpenAI’s GPT-5.6 release held up by the same pressure.
Weeks Prior to Interview
Krishnan conducts diplomatic reassurance tour of world leaders alarmed by US unilateral action on Mythos access.
July 2026 — FT Interview
Krishnan’s first in-depth post-administration interview: “There will not be an FDA for AI.” Backs equity-stake model for sharing AI upside with the American public.
The key insight: “No FDA for AI” does not mean no government control over AI. It means no published rules governing that control. The Mythos intervention and GPT-5.6 delay prove the kill switch exists. The White House is simply declining to write down when it will use it — which makes it more powerful, not less.
The Structural Read
What Krishnan described as deregulation is, structurally, something more precise: governance-by-exception replacing governance-by-statute. A licensing regime — an FDA for AI — would mean published criteria, an appeals process, predictable rules that apply equally to all labs. What the US has instead is discretionary emergency power: the ability to intervene on specific models, for national-security reasons that do not need to be publicly specified, with no procedural constraints that have been written into law.
That is not less government power over AI. That is more arbitrary power. Anthropic cannot appeal to a rulebook because there is no rulebook. OpenAI cannot predict in advance which capability thresholds will trigger a hold. The equity-stake push — encouraging labs to donate ownership stakes to the American people, an idea Krishnan says was discussed with Sam Altman despite Silicon Valley warnings it constitutes backdoor nationalisation — adds a second lever. Alongside the capability kill switch, you now have a potential ownership hook. The state shapes the frontier not through law but through leverage.
The Permission Layer — Operating Informally
“The Permission Layer is the set of forces — government, regulation, geopolitics — that determine which AI capabilities are allowed to reach users. The Trump administration has not abolished the Permission Layer. It has moved it off-statute and into the executive’s discretion. That makes it harder to challenge, harder to predict, and harder to build around.”
The geopolitical read compounds this. Krishnan’s post-departure diplomatic tour — reassuring world leaders about the Mythos intervention — reveals that unilateral US capability control is already being read by allies and rivals alike as a new instrument of foreign policy. Washington doesn’t need a treaty or a trade law. It can simply turn off a model. The precedent, not the statute, is the policy.
Three Implications
FOR ANTHROPIC AND OPENAI: Uncertainty Is Now the Regulatory Environment
Both labs now operate under a regime where the government’s intervention criteria are unpublished. Mythos was pulled; GPT-5.6 was stalled. The next trigger is unknown. This changes how frontier labs must think about capability development, release timing, and national-security disclosure — not toward a checklist, but toward constant pre-negotiation with an executive that holds a discretionary veto.
FOR NON-US LABS AND GOVERNMENTS: The US Just Demonstrated a New Power
Krishnan’s diplomatic scramble after the Mythos intervention signals that allies were not briefed and were alarmed. The episode proves the US can unilaterally restrict access to frontier models globally — no WTO mechanism, no treaty required. Every government that had integrated Mythos into its AI strategy discovered, overnight, that its access was contingent on Washington’s discretion. That is a geopolitical fact that will accelerate sovereign AI efforts in Europe, the Gulf, and Asia.
FOR THE EQUITY-STAKE PUSH: The Nationalisation Question Is Not Going Away
Krishnan’s endorsement of having AI companies donate equity to the American people — framed as ensuring voters feel the upside is shared — is politically potent and structurally ambiguous. Sam Altman has reportedly engaged with the idea. Silicon Valley reads it as backdoor nationalisation. What it actually represents is a third lever of state influence: not just the capability kill switch, not just regulatory blocking power, but a claim on ownership. If it proceeds, the US government becomes a stakeholder in the firms it is also choosing whether to regulate.
Business Engineer Framework
The Permission Layer — Inside Anthropic’s Governance Trap
The Permission Layer framework maps how government, regulation, and geopolitics function as the ultimate gatekeepers of which AI capabilities reach the market. The Mythos intervention and GPT-5.6 delay are the Permission Layer operating at maximum discretion — no statute, no checklist, just executive power. Understanding how this layer works is now a core competency for anyone building, investing in, or deploying frontier AI.
The Trump administration has not deregulated frontier AI — it has informalised its control over frontier AI. “No FDA for AI” means no published rules, no statutory process, no appeals mechanism; it does not mean no interventions. Anthropic’s Mythos is off the market. GPT-5.6 was held up. The equity-stake gambit is live. What Sriram Krishnan is describing as anti-red-tape minimalism is, in structural terms, maximum discretionary power: a kill switch with no manual, operated by an executive that also wants an ownership stake in the companies it can switch off. That is not less government. That is a different, harder-to-challenge kind of government — and every frontier lab, every allied government, and every enterprise deploying AI at scaleneeds to build its strategy around that reality, not the press line.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
