Trump Softened on Anthropic but Left Fable and Mythos Offline

President Donald Trump told Axios he no longer considers Anthropic a national security threat — a week after Commerce ordered the company to block foreign access to Fable 5 and Mythos 5, forcing a global shutdown. The warm words came after Trump met CEO Dario Amodei at the G7 summit in France. The models are still offline. The Pentagon's supply-chain ban still stands. Diplomacy moved. Policy didn't.

What Trump said: In a pretaped Axios interview published June 19, Marc Caputo asked whether Trump viewed Anthropic or Amodei as a threat.

• Trump: "Well, not now, but a week ago, maybe."
• He called Amodei "nice" and "smart" after their G7 encounter in Évian-les-Bains.
• On the June 12 export directive: Amodei responded "very quickly" and "very responsibly."
• On easing restrictions: "I would, but I'm not sure I have to do that."
• On emergency powers: Trump did not rule out the Defense Production Act but said, "I'm not sure I have to do that."
• On China: "We're beating China by a lot." He said he does not want to shut Anthropic down.

What didn't change:

• Fable 5 and Mythos 5 remain disabled for all users worldwide.
• The Commerce Department's June 12 directive has not been formally rescinded.
• The Pentagon's supply-chain risk designation from March is still in force.
• Anthropic's two lawsuits against the administration continue.

Why the gap matters: Commerce classified the models as export-controlled after concerns about a jailbreak — reportedly flagged by Amazon, Anthropic's largest investor and cloud host. Anthropic argues the technique amounts to asking a model to read code and fix flaws, capability it says rivals OpenAI's GPT-5.5. The company received the order at 5:21 p.m. ET on June 12 with no public technical detail. Because it cannot verify user nationality at inference time, compliance meant shutting down access for everyone — including U.S. citizens.

The G7 play: Amodei did not wait for a press release. At the summit he and Google DeepMind CEO Demis Hassabis pitched G7 leaders on a U.S.-led AI coalition — positioning Anthropic as a cooperative partner in American tech diplomacy, not a regulatory adversary. The strategy bought direct access to Trump at a moment the president was receptive. An Anthropic spokesperson told The Hindu: "We are grateful to the administration for their ongoing partnership in working to get this matter resolved as quickly as possible."

The Pentagon split: Defense Secretary Pete Hegseth seized on the Commerce action. "Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever," he posted on X. "Every passing day proves why that was the right move." Breaking Defense noted a community note challenging the accuracy of that claim — the February ban included a six-month phase-out, and NSA reportedly kept using a preview version of Mythos for offensive cyber operations even as the administration moved to phase out Anthropic elsewhere. Retired Lt. Gen. Jack Shanahan, former head of Project Maven, told Breaking Defense the latest order "still smacks of an ongoing vendetta against Anthropic and Dario personally" — even if Amazon's jailbreak report is taken at face value.

The IPO clock: Anthropic confidentially filed for an IPO in early June. Fortune reporting cited by The Next Web put the valuation near $965 billion. Federal restrictions cast uncertainty over the listing at the worst possible moment. Trump's conciliatory tone is a political win for bankers and investors — but export controls require bureaucratic steps a single interview cannot shortcut. Commerce operates with considerable independence. Warm words are not a product release.

Convina's view: This is what frontier AI governance looks like when there is no mature rulebook — only leverage. Trump spent a week treating Anthropic as a liability, met Amodei at the G7, and declared the threat gone while the kill switch stays pulled. That is not de-escalation. It is personalized industrial policy where the president's mood and Commerce's letter carry more weight than any transparent safety framework. Every enterprise betting on American frontier models should price two risks now: sovereign discontinuation — models switched off by export law — and sovereign rehabilitation — models restored by handshake. Neither risk appears in a standard vendor security questionnaire. Both will show up in earnings calls before the IPO season ends.