From 'left-wing nut jobs' to 'productive meeting': the White House quietly reopens its Anthropic channel

The White House described its Friday meeting with Anthropic CEO Dario Amodei as "productive and constructive" — language that lands very differently against the backdrop of the past two months, in which the administration had branded the company a "radical left, woke" supplier and ordered federal agencies to stop using its tools. The thaw, first reported by the BBC citing Axios, hints at how dependent the US government has quietly become on frontier AI from Anthropic.

The about-face in context

Amodei reportedly met with Treasury Secretary Scott Bessent and White House Chief of Staff Susie Wiles. The official readout said the conversation "explored the balance between advancing innovation and ensuring safety," and discussed shared protocols for scaling the technology.

Anthropic itself declined to comment publicly. The optics are still awkward: when President Trump originally directed government agencies to drop Anthropic, his social-media post called the company's leadership "left wing nut jobs" who were trying to "strong arm" the Department of Defense, and ended with: "We don't need it, we don't want it, and will not do business with them again!"

Asked by reporters in Phoenix about Amodei's visit, Trump said he had "no idea" about the meeting.

Why the reset is happening now

The trigger appears to be Claude Mythos, the cyber-focused model Anthropic previewed earlier this month. Anthropic claims Mythos can outperform humans on a range of hacking and computer-security tasks, and says it has surfaced bugs in decades-old code and worked out how to exploit them autonomously.

Only a few dozen organisations currently have access. Last week Amodei publicly said the company had "spoken to officials across the US government" and offered to collaborate. Friday's meeting is the first sign that those overtures are finding traction even inside an administration that, on paper, has been openly hostile.

The signal: when a model can credibly claim to be a strategic-grade cyber tool, even a politically uncomfortable supplier becomes hard to walk away from.

The lawsuit nobody dropped

The reset doesn't erase the legal fight in the background. In March, Anthropic sued the Department of Defense and other federal agencies after being labelled a "supply chain risk" — the first time a US company has publicly been given the designation, which means a technology is considered insecure for government use. Anthropic argued the label was retaliation by Defence Secretary Pete Hegseth after Amodei refused to grant the Pentagon unconstrained access to Claude over concerns about mass domestic surveillance and fully autonomous weapons.

A federal court in California has largely sided with Anthropic, but a federal appeals court declined to suspend the supply-chain-risk designation while the case proceeds. Despite that label, court records show many federal agencies that were Anthropic customers prior to the order are still using its tools.

The bigger pattern

Three things are worth flagging for anyone watching how AI policy is actually being made:

• Security utility beats political posture. Mythos has flipped the script: a White House that publicly disparaged Anthropic is now sitting across the table from its CEO because the alternative — being out of the loop on a strategic-grade cyber tool — is worse.
• The "supply chain risk" label is becoming a negotiating tool. Whatever its merits, the designation is functioning less as a final verdict and more as leverage in a long, multi-party negotiation about what unfettered government access to frontier AI should mean.
• Anthropic's positioning is now de facto bipartisan infrastructure. The combination of large-scale federal deployment, ongoing court wins, and a model the Treasury wants major banks to test against, makes the company harder to dislodge with a single executive order.

What to watch next

The immediate question is whether Friday's meeting produces concrete commitments — for example, formalised access to Mythos for federal cyber-defence teams, or any softening of the supply-chain designation. The deeper question is structural: how does any administration, of any party, manage a vendor whose product is simultaneously a national-security asset and a political flashpoint?

Mythos is forcing that question into the open faster than most observers expected. Friday's "productive" meeting suggests the answer, at least for now, is pragmatism over posture.

(Source: BBC News. Photo: Unsplash, licence.)