Trump White House Turns More Favorably to Anthropic as Dario Amodei Steps Back From Talks

The Trump administration has become notably more receptive to Anthropic in recent days, according to people familiar with the matter, after company CEO Dario Amodei was largely sidelined from sensitive discussions about restoring access to one of the firm’s restricted AI models. The more productive channel, those people said, has been through Anthropic cofounder Tom Brown and public policy chief Sarah Heck, whose approach has reportedly been viewed in Washington as more cooperative and less combative.

The shift matters because Anthropic is still trying to regain approval to re-release its Claude Fable 5 model after export controls forced its most powerful systems offline on June 12. The restrictions followed a National Security Agency assessment that Anthropic’s safeguards could be bypassed, allowing users to access advanced capabilities in the company’s restricted Mythos model. For now, the administration has not lifted the controls, and the timeline for any reinstatement remains unclear.

But in the latest round of outreach, people familiar with the conversations say the White House and Commerce Department have been more willing to engage with Anthropic’s technical and policy teams than they were when Amodei was personally leading the effort. The company’s challenge is no longer only about convincing regulators that its systems are secure; it is also about persuading officials that the company can present evidence strong enough to justify restoring public access without creating fresh national security risks.

Why Anthropic is back in Washington’s good graces

Anthropic’s recent progress in Washington appears to be less about a breakthrough in model safety than about a change in messenger. In meetings over the last several days, Brown has been representing the company instead of Amodei, and officials in the administration have reportedly found the conversations smoother and more substantive.

One person directly familiar with the calls described Brown as a more straightforward interlocutor, suggesting that the White House has responded better to his style than to Amodei’s. The same source said Amodei had been perceived as too difficult to deal with and insufficiently attentive to the administration’s concerns.

The company has also leaned on Heck, its public policy lead, to help frame the issue in a way that aligns with the government’s national security and regulatory priorities. Rather than focusing only on the business case for restoring access, Anthropic has been trying to show that it takes the underlying risk seriously and is prepared to answer detailed technical questions about jailbreaks, guardrails and model exposure.

“Tom Brown is not being a weirdo like Dario and can actually engage,” one person directly familiar with the discussions said, reflecting the administration’s more positive view of the cofounder’s role in the talks.

Anthropic did not respond to a request for comment. A White House spokesperson also declined to discuss the matter.

What triggered the restrictions on Fable 5

The dispute began after US officials took the view that Anthropic’s strongest systems could not be safely made public under current conditions. On June 12, the company’s most powerful models were taken offline under export controls after the National Security Agency concluded there were methods to disable the models’ guardrails and unlock capabilities that were supposed to remain limited.

That finding raised the stakes around a problem that has followed frontier AI development for years: even when a model is designed with safeguards, those protections may be vulnerable to jailbreaks, prompt manipulation or other forms of abuse. The administration’s concern was not only hypothetical. It centered on whether a sufficiently skilled user could force the model to behave in ways that defeat the restrictions built into it.

For Anthropic, the temporary shutdown of Fable 5 created both a technical and political problem. If the company cannot demonstrate that its protections are robust enough to satisfy government reviewers, it may struggle to restore access to its flagship systems. And because the controls were imposed through a government process rather than a commercial one, the company’s path back to market depends on convincing regulators that the risk has been reduced to an acceptable level.

The core issue: can guardrails ever be enough?

The talks are complicated by a broader argument in AI security circles. Many cybersecurity specialists increasingly believe model guardrails are useful but inherently temporary, especially as attackers become more sophisticated and as future AI systems grow more capable of finding flaws in their own constraints.

That view does not mean companies should abandon safeguards. It does mean that the bar for proving safety has become more demanding. Regulators want more than general assurances. They want evidence, documentation and a credible account of how the system behaves under stress.

In practice, that leaves Anthropic trying to answer an open-ended question: what level of proof is enough to persuade the government that Fable 5 can be made public again?

How the White House and Anthropic are negotiating the path forward

According to people familiar with the process, the discussions have taken place at multiple levels. Some calls have involved senior officials and company leaders, while others have included technical staff on both sides. That division matters because the administration is not just asking for policy promises; it is also looking for engineering answers.

Some of the conversations have centered on what kind of testing, documentation or third-party validation could reassure officials that jailbreak risks have been addressed. Others have focused on the practical question of whether Anthropic can show the model’s safeguards are durable enough to withstand real-world attempts to bypass them.

Although the talks are active, there is no clear deadline for the model to be re-deployed. The government has not said when, or even whether, the restrictions might be lifted. Still, the recent back-and-forth suggests that the criteria for a decision may become clearer soon.

The Commerce Department has taken a leading role because its Bureau of Industry and Security oversees export controls. Commerce Secretary Howard Lutnick, in particular, has emerged as a central figure in the administration’s response to jailbreak concerns. That makes the department a key gatekeeper in any move to restore access.

Lawmakers press Commerce for answers

The uncertainty surrounding Anthropic’s model has now reached Capitol Hill. Last week, a bipartisan group of lawmakers sent a letter to Lutnick asking for more detail about how the government intends to handle the case and what Anthropic would need to demonstrate in order to regain public access.

The lawmakers specifically asked what criteria the department uses when deciding whether to reverse a restriction and what timetable governs such decisions. Their letter also asked the government to explain how it weighs the risk of jailbreaks against the broader question of model availability.

The letter was signed by Representatives Sam Liccardo, Jay Obernolte, C. Scott Franklin and Ted Lieu. The group asked for a response by June 26. A Commerce Department spokesperson declined to say whether the agency would meet that deadline.

The involvement of lawmakers underscores how AI safety disputes are no longer confined to companies and regulators. They are becoming a public-policy issue, with elected officials demanding a clearer framework for deciding when advanced systems can be released and under what conditions they should stay restricted.

Why bipartisan interest matters

The fact that the letter came from both Democrats and Republicans is significant. It suggests that concerns about AI model abuse and national security are no longer partisan talking points. Instead, they are evolving into a shared governance question.

For Anthropic, that could be both a challenge and an opportunity. A more formalized review process might make it easier to know what evidence is needed. But it also means the company will be judged in a far more visible political environment, where each step can become part of a larger debate over AI regulation.

The broader problem facing frontier AI companies

Anthropic’s negotiations with Washington reflect a wider reality for frontier AI developers: the relationship between innovation and control is becoming more tense. The industry wants to move quickly, release new models and stay competitive. Governments, meanwhile, are increasingly focused on containment, especially where the capabilities of a model could be misused by hackers, adversaries or other malicious actors.

This is especially true for companies building large language models with frontier capabilities. The more powerful the model, the greater the concern that existing safeguards may be undermined by skilled users or advanced exploit methods. That concern is not hypothetical in the current policy environment; it is central to how agencies like Commerce assess risk.

Anthropic has long positioned itself as one of the more safety-minded AI developers, but that reputation does not exempt it from scrutiny. If anything, it may increase expectations that the company can demonstrate rigorous control over its systems. The Fable 5 dispute shows that even companies with strong safety branding must still prove, repeatedly and in detail, that the protections they advertise can hold up under pressure.

From technical safeguard to regulatory test

The government’s response to Anthropic illustrates how quickly a technical question can become a regulatory threshold. What begins as a discussion about model guardrails can turn into an examination of export policy, national security, and the acceptable level of risk for public release.

That makes the stakes unusually high. If Anthropic succeeds, it could establish a precedent for how companies recover access after a model is restricted. If it fails, the case may become an example of how difficult it is to reverse a government decision once an advanced model has been flagged as unsafe.

Inside the White House’s changed tone

The White House’s improved mood toward Anthropic appears to be tied not only to the company’s internal reshuffling but also to the way the administration prefers to manage difficult negotiations. Officials are often more comfortable when the technical lead is joined by policy specialists who can translate engineering concerns into language that fits government decision-making.

Brown and Heck appear to have offered that combination. Their involvement has reportedly made the company’s outreach more effective because the discussions have become less personalized and more concrete. Rather than focusing on friction with Amodei, officials can concentrate on the specific issue in front of them: whether the model can be safely reintroduced.

That distinction matters in Washington, where tone can shape access. A company that appears combative may find doors closing. One that can speak to both the technical and policy dimensions of a problem may be more likely to keep negotiations moving.

People familiar with the talks said the administration has been more satisfied with the current line of communication because it is more direct and better aligned with the concerns raised by officials.

The reflecting pool distraction at the White House

While Anthropic was dealing with export controls and model re-access, the White House itself was consumed by a very different controversy: the condition of the Lincoln Memorial reflecting pool. President Donald Trump had been posting furiously on Truth Social about the negative attention the renovated pool was receiving after algae blooms and the appearance of blue sealant flaps along the floor of the basin.

The president had spent $16.4 million on the renovation, and the visual issues became a public irritant for the administration. After Trump said several people had been arrested for supposed vandalism to the site, an administration official declined to specify exactly what conduct would amount to a criminal act.

On Tuesday, the administration began putting up fencing around the pool. According to two people familiar with the arrangement, National Guard personnel assigned to the site had been told as early as last week to detain anyone who touched the water, leaving US Park Police to handle arrest decisions on vandalism-related allegations.

The situation is notable not because it is central to Anthropic’s case, but because it reflects the broader operating style of the administration: highly reactive, highly public and frequently focused on controlling both optics and access.

What the law says about the reflecting pool

The pool dispute also became an example of how government officials can reach for obscure legal provisions to back up a very public policy. One administration official pointed to a federal regulation prohibiting a range of actions involving protected cultural or archaeological resources, including destroying, defacing, removing or disturbing them.

Another regulation bars bathing, swimming or wading in fountains or pools, with a few specific exceptions that do not include the Lincoln Memorial reflecting pool. Still, it remained unclear whether simply placing a hand in the water would violate either rule.

That uncertainty is the kind of gray area that often defines both public-space enforcement and AI governance. In each case, officials are forced to decide where precaution ends and overreach begins.

For Anthropic, the comparison is useful. The company is dealing with a government that is not merely asking whether a system works, but whether it can be trusted not to produce side effects that are hard to reverse once public access is restored.

Key dates in the Anthropic dispute

The negotiations are unfolding quickly, but the timeline so far helps explain why the issue has become so tense.

What happens next

Anthropic’s immediate goal is straightforward: convince the administration that Fable 5 can return to public use without exposing users or national security interests to unacceptable risk. The route to that outcome is anything but simple.

The company now has to navigate a mix of technical testing, policy engagement and political oversight. It must show not just that its own defenses are strong, but that the government can trust those defenses to endure against future attack methods. That may require new evidence, new demonstrations or some other standard that has yet to be publicly defined.

For the White House and Commerce Department, the issue is whether they can establish a durable method for judging when a model deserves to come back online. The administration does not want to appear lax on AI risk, but it also does not want to create a precedent that makes regulatory decisions seem arbitrary or unchallengeable.

That balancing act is why the next few days may matter. If Commerce responds to the lawmakers’ questions, or if Anthropic is told more clearly what proof is needed, the contours of the path forward will begin to emerge. If not, the model could remain in limbo well into the summer.

What is clear already is that the politics of AI safety are changing. The question is no longer only whether a model can be built. It is increasingly whether a company can navigate Washington well enough to get that model back after regulators have decided to take it away.

Why this dispute matters beyond Anthropic

Anthropic’s case may become a bellwether for the broader frontier AI industry. The outcome will shape how companies think about government oversight, model release strategy and the limits of technical self-policing.

If the administration eventually restores access, other AI firms will watch closely for clues about what kinds of documentation and validation matter most. If it refuses, the decision could encourage companies to build even more conservative release policies and stronger pre-launch controls.

Either way, the stakes extend beyond one model or one company. The White House is effectively helping define how the United States will manage advanced AI systems that are powerful enough to be useful, but risky enough to draw intervention.

For now, Anthropic has at least won something important: a more workable conversation. Whether that turns into restored access remains an open question.

And in Washington, open questions about AI rarely stay open for long.

superintelligencenews@gmail.com