Anthropic White House Talks Back on Track After Amodei Steps Aside

The Anthropic White House talks that have been quietly grinding away for nearly two weeks appear to have found a new gear — largely because the person sitting across from administration officials changed. CEO Dario Amodei is out of the room. Co-founder Tom Brown is in. And by all accounts, that swap has made a material difference.

• Anthropic White House talks stalled under CEO Dario Amodei, who sources described as difficult to engage with.
• Co-founder Tom Brown now leads Anthropic White House talks and is reportedly making meaningful progress with the administration.
• Claude Fable 5 was pulled offline on June 12 after Amazon researchers flagged jailbreaks that could bypass its safety restrictions.
• The White House had already grown concerned that actors linked to Chinese interests had accessed Fable 5 before the jailbreak issue emerged.

How the Anthropic White House Talks Went Off the Rails

When Anthropic first entered negotiations with the Trump administration over a sweeping export control directive, Amodei was front and center. That, according to sources cited by Wired, was a problem. The picture that’s emerged from those early sessions isn’t flattering: Amodei reportedly struggled to listen, couldn’t stay on track, and left administration officials frustrated. One anonymous participant in the calls reportedly described Amodei in unflattering terms, contrasting him with Brown, who could actually engage.

That’s a brutal characterisation of a sitting CEO, especially one whose company is currently trying to win a regulatory favour from the most powerful executive branch in the world. But it also isn’t entirely out of left field. Previous reporting has painted Amodei as prone to emotional outbursts and extended monologues. Anyone who’s watched him speak publicly — even in friendly, interview-style settings — will recognise the particular energy: the sweeping arm gestures, the downward head tilts mid-sentence, the trumpet-register voice. He’s clearly brilliant. He’s not always easy to read across a table.

To be clear, none of this is disqualifying for running an AI lab. Some of the most consequential technology executives in history have had famously difficult interpersonal styles. But when you need a government regulator to lift a restriction that’s actively harming your business, the ability to sit, listen, and build rapport suddenly matters a great deal more than it does in a boardroom or on a conference stage. The Anthropic White House talks exposed exactly that gap.

Tom Brown Steps In — and the Mood Shifts

Brown’s LinkedIn lists his title simply as ‘co-founder,’ which is either a sign of supreme confidence or supreme indifference to personal branding — possibly both. What matters is that his entry into the Anthropic White House talks appears to have changed the atmosphere almost immediately. He’s joined at the table by Sarah Heck, Anthropic’s Head of Public Policy, whose role almost by definition requires her to be measured, on-message, and strategically disciplined. That’s exactly the combination you’d want in a negotiation like this: someone with technical founder credibility paired with someone who knows how to navigate government relations without detonating the conversation.

From what’s publicly available of Brown online, he comes across as warmer and more conventionally personable than Amodei — softer-spoken, more comfortable making sustained eye contact, quicker to smile. That might sound like an odd thing to focus on in a story about AI regulation, but in a negotiation where personal dynamics have apparently been a deciding factor, it’s genuinely relevant context.

What Anthropic Is Actually Trying to Get Done

The stakes here are significant. Anthropic needs the White House to clear Claude Fable 5 — its most capable consumer-facing model — for general use again. Fable 5 was pulled offline on June 12, a move Anthropic announced publicly after the administration’s export control directive made it untenable to keep the model running without restrictions. Observers tracking the Anthropic White House talks closely say this single issue — restoring Fable 5 — remains the central sticking point in every session.

Anthropic classifies Fable 5 as a ‘Mythos-class’ model, meaning it shares foundational technology with Claude Mythos Preview — a model the company itself determined was too dangerous for open public access. The Fable 5 version was supposed to have the sharpest edges sanded off: capabilities that could pose a cybersecurity risk were meant to have been stripped out before release.

That assurance didn’t hold up for long. According to multiple sources, researchers at Amazon — which has a major investment stake in Anthropic — notified the White House shortly after Fable 5 launched that jailbreaks could apparently remove those safety guardrails with relative ease. If that’s accurate, it means Anthropic shipped a model it described as safely restricted, only for third-party researchers to immediately poke holes in those restrictions. That’s not a good look, and it almost certainly hardened the administration’s posture going into these negotiations.

Chinese Access Concerns Came First

What makes the timeline particularly uncomfortable for Anthropic is that the White House was reportedly already alarmed before the jailbreak findings even surfaced. Officials had grown concerned that individuals with ties to Chinese interests had gained access to Fable 5 in the brief window before controls were put in place. The export control order — which required Anthropic to block non-U.S. nationals from using both Fable 5 and the more restricted Mythos 5 — was issued approximately three days after the model’s launch. That’s an extraordinarily fast regulatory response, which tells you how seriously the administration was taking the threat.

The geopolitical dimension here is worth sitting with for a moment. The U.S. government’s anxiety about advanced AI capabilities ending up in Chinese hands isn’t new — it’s been a running theme through chip export controls, restrictions on NVIDIA hardware sales, and ongoing scrutiny of research partnerships between American universities and Chinese institutions. Anthropic is now caught in the middle of that broader tension, and the Anthropic White House talks are as much about national security posture as they are about any one company’s product roadmap.

Anthropic White House Talks: What a Resolution Might Look Like

If the Anthropic White House talks do reach a workable agreement, it almost certainly won’t be a clean, unconditional restoration of Fable 5’s availability. More likely, any resolution involves Anthropic committing to specific technical safeguards, agreeing to third-party auditing of its safety restrictions, or accepting some ongoing monitoring mechanism as a condition of getting the model back online. The administration isn’t going to simply take Anthropic’s word for it again — not after the jailbreak episode.

That actually puts Anthropic in an interesting structural position relative to its competitors. If it has to build more rigorous, externally verifiable safety frameworks as a condition of this deal, those frameworks could ultimately become a selling point — evidence that its models have been stress-tested under adversarial conditions in a way that OpenAI’s or Google DeepMind’s haven’t been forced to demonstrate publicly. There’s a version of this story where Anthropic comes out of a bruising regulatory fight with a credibility advantage.

Whether Brown and Heck can actually land that outcome remains to be seen. But the early signs — at least according to the people in the room — suggest the conversation has finally moved past the point where personal dynamics were getting in the way of actual progress. In a negotiation this high-stakes, that’s not a minor thing. That’s the whole game.

Source: Gizmodo

Frequently Asked Questions