The White House Stopped Talking to Amodei. That Tells You Everything.

The Trump administration has stopped dealing with Anthropic CEO Dario Amodei. In his place, cofounder Tom Brown is now leading high-stakes negotiations about re-releasing the Claude Fable 5 AI model. A White House source told WIRED: “Tom Brown is not being a weirdo like Dario and can actually engage.” The administration has not lifted export controls that took Anthropic’s most powerful models offline on June 12 — nearly two weeks ago.

🔍 THE BOTTOM LINE

The replacement of Amodei with Brown signals that Washington views this as a negotiation, not a safety review. The administration wants a partner who complies, not one who raises safety concerns. For every nation — including New Zealand — that depends on US frontier models, the message is clear: access to the most powerful AI systems is now contingent on political alignment, not technical readiness.

What “Too Difficult” Actually Means

The WIRED reporting is blunt. Amodei was “too difficult to talk to and did not listen to their concerns.” Brown and Anthropic’s public policy chief Sarah Heck are now leading outreach. Multiple calls have occurred in recent days, at both the political and technical working-group levels.

This isn’t a personality clash. It’s a structural disagreement about who controls AI safety decisions. The June 12 export control directive was triggered by the NSA confirming that guardrails on Anthropic’s Mythos model could be bypassed. Amodei’s position — that safety guardrails are non-negotiable and that the government should not dictate internal model architecture — is the stance of a company that built its brand on responsible AI. Brown’s appointment suggests the administration found that position inconvenient.

The contrast with the G7 meeting just days ago is stark. At Evian-les-Bains, Trump said Anthropic had “behaved very responsibly.” Now his administration is sidelining the CEO who insisted on those responsible behaviors.

The June 26 Deadline

Bipartisan lawmakers — Representatives Sam Liccardo, Jay Obernolte, C. Scott Franklin, and Ted Lieu — have sent a letter to Commerce Secretary Howard Lutnick demanding answers by June 26. Among the questions: “What specific criteria does the Department rely upon for determining whether to restore public access to the model through a revision of this decision? What is the timeline for that decision?”

Lutnick has taken a leading role because the Commerce Department’s Bureau of Industry and Security manages export controls. A Commerce Department spokesperson declined to comment on whether the agency would respond by the deadline.

The talks now involve technical staff from both sides, attempting to establish what level of proof from Anthropic might alleviate concerns about Fable 5 jailbreaks. But the underlying tension — whether guardrails on AI models are a stopgap or a permanent solution — remains unresolved. As Inner Loop has noted, independent cybersecurity experts increasingly view model guardrails as temporary, since skilled users and future AI models will eventually find ways to bypass them.

The Mythos Precedent

The Mythos Discord breach earlier this year proved that determined outsiders don’t need Washington’s permission to access restricted models. The administration’s concern about jailbreaks is not abstract — it’s based on demonstrated failures. But the response — restricting the model entirely and demanding Anthropic prove a negative (that no jailbreak is possible) — sets a precedent that could keep any frontier model permanently offline.

This is the structural problem Amodei was reportedly resisting. You cannot prove a model is unjailbreakable, because the attack surface evolves with every new technique. Demanding that proof as a precondition for redeployment is not a safety review — it’s a veto.

NZ Angle — The Dependency Problem

For New Zealand’s tech sector, this is a case study in why dependency on US frontier models is a strategic risk. NZ developers and researchers rely on access to models like Fable 5 for everything from scientific research to enterprise applications. If access is contingent on Washington’s political satisfaction with the CEO’s personality — not the model’s safety profile — then NZ’s AI capability is effectively controlled by US domestic politics.

Anthropic employees being targeted by the administration adds another dimension. If key researchers leave or are constrained, the model development pipeline itself degrades — and NZ’s access to future frontier models degrades with it. The EFF’s First Amendment challenge to the export controls is the only legal path currently active, but it moves slowly.

The Other Side

Critics argue that the administration’s concerns are legitimate — the NSA did confirm that Mythos guardrails could be bypassed, and Fable 5 is a frontier model with significant capability. The export controls weren’t arbitrary; they were a response to a verified security failure. And Brown’s engagement may simply be more productive than Amodei’s resistance, leading to faster resolution rather than weaker safety standards.

But this framing ignores the core issue: the administration is not asking Anthropic to prove Fable 5 is safe. It’s asking Anthropic to prove Fable 5 is unjailbreakable — which is not the same thing. No frontier model has ever met that standard, and none ever will.

❓ FAQ

Q: Why was Dario Amodei replaced in the negotiations? A: White House sources told WIRED that Amodei was “too difficult to talk to” and “did not listen to their concerns.” Cofounder Tom Brown is viewed as more cooperative. The replacement suggests the administration prioritizes compliance over technical debate.

Q: What triggered the export controls on June 12? A: The NSA confirmed that guardrails on Anthropic’s Mythos model could be disabled or bypassed, which the Commerce Department treated as a national security risk requiring export control restrictions on Anthropic’s most powerful models.

Q: Will Fable 5 be redeployed soon? A: The timeline remains uncertain. Bipartisan lawmakers have demanded criteria for redeployment from Commerce Secretary Lutnick by June 26, but the Commerce Department has not confirmed it will respond by that deadline.

Q: Does this affect NZ companies using Anthropic’s API? A: Directly, not yet — standard API access for commercial use hasn’t been explicitly restricted. But the precedent — where political friction with a CEO can take frontier models offline for weeks — creates investment risk for any NZ business building on Anthropic’s platform.

Q: What’s the difference between proving a model is “safe” and proving it’s “unjailbreakable”? A: “Safe” means the model’s guardrails work as intended under normal use. “Unjailbreakable” means no possible technique can bypass those guardrails — a standard no frontier model has ever met. The administration appears to be demanding the latter, which effectively means no frontier model can ever be fully unrestricted.

🔍 THE BOTTOM LINE

The White House replacing Amodei with Brown is not a personnel decision — it’s a signal that access to frontier AI is now a political negotiation, not a technical review. For New Zealand and every other nation dependent on US AI exports, the lesson is the same: build resilience against the possibility that the model you depend on tomorrow can be taken offline by a personality clash in Washington today.