For three months, Anthropic held the unusual distinction of being the only American AI company officially classified as a national security supply-chain risk by its own government. That designation, handed down by the Defense Department in March, barred most federal agencies from contracting with the company even as Claude Mythos was quietly solving cybersecurity problems the government itself could not. The standoff is softening, but only at one end.
Multiple people familiar with the discussions told Reuters that Anthropic has made meaningful progress in its relationship with the White House, even as Defense Secretary Pete Hegseth continues to reject the company's request to reconsider the blacklist. The divergence reflects a broader split inside the Trump administration about how to handle Anthropic, and it matters considerably as the company prepares an initial public offering that could set a valuation record for an AI startup. Anthropic filed a confidential S-1 prospectus with the SEC on June 1, with bankers targeting a late-2026 debut at a valuation approaching $1 trillion.
The Fault Lines Within the Administration
The rupture began when Anthropic refused to allow the US military to use Claude models for domestic surveillance programs or fully autonomous weapons systems. The refusal was principled; it was also politically costly. The Defense Department's response came in March: a supply-chain risk designation, the first ever applied to an American company in the AI sector, that effectively locked Anthropic out of most federal procurement channels.
But not everyone in Washington drew the same conclusions. Dario Amodei visited the White House in mid-April for the first face-to-face meeting between Anthropic's chief executive and administration officials since the dispute began. That visit set off a series of quieter conversations. Anthropic employees met with Treasury Secretary Scott Bessent to discuss Mythos and potential executive actions on AI. Separately, discussions with National Cyber Director Sean Cairncross covered how Mythos could be deployed to protect critical infrastructure from AI-enabled cyberattacks. Those conversations informed, at least in part, the executive order on AI that the administration published on June 2.
Key Facts
- DoD designationMarch 2026, first US AI company labeled a supply-chain risk
- First White House meetingApril 2026, Dario Amodei's first visit since dispute began
- Treasury briefingAnthropic staff met Scott Bessent on Mythos & executive AI actions
- Executive order linkJune 2 AI order partly informed by Anthropic-White House talks
- Pentagon stancePete Hegseth denied Anthropic's request to reconsider designation
- NSA exceptionIntelligence agency separately gained Mythos access despite DoD blacklist
A Split Bureaucracy
The Hill reported that the White House and the Pentagon have drifted apart in how they handle Anthropic, with the executive office moving toward accommodation and the Defense Department refusing to budge. That split has real-world consequences. The NSA, which operates with some independence from the Pentagon's procurement hierarchy, has already obtained access to Mythos. The State Department and parts of the intelligence community appear to be drawing their own conclusions about the model's value, regardless of what the DoD designation formally says.
Nextgov reported in April that the White House was drafting a framework to permit federal use of Anthropic's models under specific conditions. If finalized, that framework would create an official channel for civilian agencies to use Claude even while the DoD designation technically remains active. Whether Hegseth accepts such a workaround, or whether the Pentagon eventually moves to formally rescind the designation, is not yet clear. Anthropic's legal challenge to the blacklist has not produced a ruling, and political negotiations have so far advanced further than litigation.
"The White House and the Pentagon are taking significantly different approaches to how, and whether, the federal government uses Anthropic's artificial intelligence systems, as the White House warms up to the frontier AI company and the Pentagon digs its heels in." The Hill, June 5, 2026
What Comes Next for the IPO
The uncertainty arrives at a consequential moment. An IPO roadshow is considerably harder to run when a government classification lets investors argue the company is a security liability. Anthropic's underwriters at Morgan Stanley and Goldman Sachs will need a clean answer about federal standing before the company begins meeting with institutional investors, likely in the autumn.
The trajectory of events, for now, favors Anthropic. The company's relationship with the NSA and the discussions with Treasury and the National Cyber Director suggest that the national security community views Claude Mythos as an asset rather than a liability, regardless of the DoD label. Whether that informal consensus translates into a formal status change before the IPO window opens is the question that will shape how aggressively Anthropic can market itself to public investors.
The underlying dispute, over autonomous weapons and domestic AI surveillance, has not been resolved. Anthropic's refusal to allow those use cases remains in place. What has changed is the White House's willingness to treat that refusal as a negotiating position rather than a disqualifying stance. The Pentagon's position, under Hegseth, has not moved. The question now is whether the White House eventually pulls rank, or whether the two sides of the administration simply proceed on parallel tracks.
