The White House Turned Off Fable 5, Who’s Really in Control of America’s AI?

Anthropic's Fable and Mythos AI models remain unavailable days after a federal export control directive forced the company to suspend access, even as the White House confirms negotiations with the company are underway.

The shutdown may have begun with a reported jailbreak. It has since become a test of how much authority the federal government can exercise over advanced AI products once they reach the market.

On Friday, June 12, Anthropic announced that a federal directive had forced the company to suspend access to Fable 5 and Mythos 5 after government officials raised concerns about a reported jailbreak affecting the system. The move effectively took two of the world's most advanced AI models offline and immediately sparked debate across the technology industry. Days later, Trump administration officials appeared to shift their public messaging, suggesting that concerns may also involve the possibility that China could gain access to the model.

The models remain unavailable.

The most significant development since the shutdown is not a restoration of access. It is the confirmation that Anthropic and the federal government are actively negotiating behind the scenes.

On June 17 at the G7 meeting, President Donald Trump told reporters that discussions with the company were "going fine," providing the first public acknowledgment that talks were underway. What remains unknown is what those negotiations are intended to achieve, what conditions Anthropic may be expected to meet, or what standard officials will use to determine whether access can be restored.

That uncertainty has become a central part of the story.

The federal government has yet to publicly release detailed evidence supporting its original decision.

Anthropic has argued that the issue involved a limited jailbreak rather than a broader failure of the model's safety protections. According to the company, the reported vulnerability could identify a small number of software flaws but did not represent a universal method for bypassing safeguards.

Axios White House reporter Sophia Cai reported that government concerns centered on a jailbreak affecting the model. Anthropic disputes the severity of the issue, describing it as a non-universal jailbreak.

That distinction may sound technical, but it sits at the center of the dispute.

I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:

— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.…

— David Sacks (@DavidSacks) June 13, 2026

Without more public information, it remains difficult to determine whether the government's response was proportionate to the threat it identified.

Complicating matters further, administration officials have increasingly begun emphasizing a different concern than the one initially described. While the original public explanation focused on the reported jailbreak, some officials have more recently highlighted the possibility of Chinese infiltration, access, or exploitation of the system. That narrative was not central to the government's initial public justification for the shutdown, and officials have not fully explained how it relates to the reported jailbreak or whether it represents a separate concern altogether.

The emergence of that argument has fueled growing anxiety inside Anthropic. According to reporting by The New York Times, some employees have begun to suspect that the administration's objectives extend beyond addressing a specific security vulnerability. Internally, a number of employees reportedly fear that the federal government may ultimately be seeking to cripple or shut down Anthropic itself rather than simply resolve concerns surrounding Fable 5.

Whether those fears are justified remains unclear. But the widening gap between the government's evolving public explanations and Anthropic's characterization of the incident has deepened mistrust between the two sides.

Negotiations Raise New Questions

The existence of negotiations creates a new layer of uncertainty. At this point, the public does not know what is going on.

The administration has not publicly identified specific conditions Anthropic must satisfy before access can be restored. Anthropic has not disclosed whether it has agreed to technical changes, additional restrictions, or other concessions.

As a result, the public and major corporate clients are left with an unusual situation. A frontier AI model has been removed from the market. Negotiations are taking place behind closed doors. Neither side has fully explained what standard will determine when access returns.

The Fable 5 Disruption Exposed a Different Kind of Risk

Companies are increasingly building products, internal tools, customer-service systems, and research efforts around frontier AI models.

Most spend their time worrying about technical failures, cyberattacks, or service outages.

Few expected a newly released model to become unavailable because of a federal directive.

Businesses can usually adapt to changing security requirements or regulatory rules. What is harder to manage is uncertainty. If a government can arbitrarily suspend access to an AI model after release, companies need to know what standards trigger that intervention and how to recover from such a shutdown.

Right now, those answers remain largely out of public view.

Anthropic's National Security Dilemma Arrived Amid Existing Controversy

Before federal officials intervened, Anthropic was already facing criticism over a 30-day data-retention requirement. The company also faced criticism for policies that could restrict or downgrade users based on how it interpreted activity on the platform.

Critics argued that Anthropic was positioning itself as a gatekeeper, deciding who should receive access to its most advanced capabilities.

Supporters viewed those policies as a reasonable response to increasingly powerful technology.

Then the federal government imposed a far broader restriction than anything Anthropic itself had implemented.

That comparison reshaped the conversation.

For years, public debate has focused largely on whether AI companies possess too much control over advanced systems.

The Fable episode added another layer to that discussion.

How much authority should governments have once those systems are released?

No comprehensive federal law specifically regulates large language models. Oversight remains fragmented, relying largely on executive actions and national-security export rules.

If regulators intervene whenever a significant jailbreak is discovered, companies will need to understand where that threshold sits. At the moment, those standards remain largely hidden from public view.

For now, the shutdown continues.

Negotiations are underway. Access remains restricted. The standards guiding the government's decision remain unclear.

That uncertainty may be the most important part of the story.

The question is no longer whether the federal government can intervene after a frontier AI model is released.

It already has.

The question now is what happens next.

For years, public debate has focused largely on whether AI companies possess too much control over advanced systems.

The Fable episode added another layer to that discussion.

How much authority should governments have once those systems are released? No comprehensive federal law specifically regulates large language models. Oversight remains fragmented, relying largely on executive actions and national-security export rules.

If regulators intervene whenever a significant jailbreak is discovered, companies will need to understand where that threshold sits. At the moment, those standards remain largely out of public view.

The models are back online and the immediate crisis may be over.

But something important happened over those few days. The US federal government demonstrated that it will arbitrarily intervene after a model is released.

Until now, that possibility was largely theoretical.

Now, it is part of doing business.