In short
Anthropic’s Mythos 5 has been cleared for a limited set of approved users after two weeks of talks with the Trump administration. Fable 5 remains restricted, leaving the company’s broader rollout plans unresolved.
- The U.S. government is allowing limited access to Mythos 5 for approved cyber defenders and infrastructure providers.
- Anthropic’s public-facing Fable 5 model remains restricted with no timetable for release.
- The administration kept the broader June 12 restrictions in place, but added a narrow exception.
- OpenAI has received a similar limited-access arrangement for GPT-5.6, hinting at a broader policy pattern.
Anthropic’s latest regulatory fight with the Trump administration has taken another turn: the company’s Mythos 5 model is being restored for a limited set of approved users, but the public-facing Fable 5 system remains unavailable as negotiations continue. The shift, confirmed in a June 26 letter from Commerce Secretary Howard Lutnick to Anthropic co-founder Tom Brown and viewed by The Verge, signals a partial easing of restrictions without a full rollback of the government’s earlier directive.
The decision gives Anthropic a path to resume access to Mythos 5 for a small group of cyber defenders and infrastructure providers, including some non-U.S. nationals inside those approved organizations. But the government has kept the broader controls in place, leaving the company’s public rollout plans for Fable 5 unresolved and its long-term release strategy dependent on further talks with Washington.
What changed in the latest government letter
The June 26 notice does not repeal the export-control style restrictions imposed earlier this month. Instead, it revises the licensing conditions enough to carve out a narrow exception for Mythos 5, Anthropic’s cybersecurity-focused model. The change follows two weeks of tense negotiations between the company and federal officials, during which Anthropic sought relief from restrictions that had blocked access even for some of its own employees who are not U.S. nationals.
Lutnick said in the letter that the company had worked with the U.S. government to reduce risks tied to both Mythos 5 and Fable 5, and that those efforts had produced enough progress to justify allowing “trusted partners” to use the model again. The letter also says Anthropic has committed to work with the government on protocols, standards and release procedures for future Mythos-class systems.
“In light of this progress,” Lutnick wrote, “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.”
That wording matters. The administration is not saying the model is now broadly safe for unrestricted release. It is saying that a specific subset of organizations can be trusted to use it under a narrower framework while the larger policy debate continues.
Mythos 5 returns, but only to a small circle
According to Anthropic spokesperson Danielle Ghiglieri, the company received formal notice that Mythos 5, which she described as its strongest cybersecurity model, can be redeployed to a limited group of cyber defenders and infrastructure providers. Anthropic says it is now working to provision those approved users and restore access as quickly as possible.
Ghiglieri said the company welcomes the move and is continuing to press for broader access, including eventual general availability of Fable 5.
That distinction is crucial. Mythos 5 is the specialized model at the center of the current dispute, while Fable 5 is the public-facing version that Anthropic wants to bring to general users. The government’s latest action only addresses the first of those two products, leaving the second one in limbo with no set timeline for release.
For Anthropic, the practical effect is partial restoration rather than a full reopening. The company can begin serving approved organizations again, but it still cannot treat Mythos 5 or Fable 5 as normal commercial offerings available on Anthropic’s usual terms.
The broader restriction remains in force
One of the most important details in the letter is what did not change. The Trump administration did not lift the underlying directive issued two weeks ago, which had barred foreign nationals from accessing either model. That prohibition was unusually broad, reaching beyond end users and affecting Anthropic staff as well.
Under the new exception, access for Mythos 5 is now allowed for approved organizations and the non-U.S. nationals who work for them, as well as non-U.S. national Anthropic employees who need to interact with the model for deployment and support. But the company’s overall situation remains constrained by the original order, which is still active until further notice.
Lutnick made that point explicit, writing that all other requirements from the June 12 letter remain in force. He also reserved the government’s right to alter the scope again if conditions change.
This means Anthropic’s reprieve could prove temporary or incomplete. The administration is still preserving wide latitude to tighten or loosen the rules as it sees fit, rather than creating a permanent pathway for general model release.
Why the administration appears to have relented, at least partly
The rollback-by-exception approach reflects growing pressure on the White House from several directions. Inside the U.S. AI industry, there has been increasing concern that highly selective restrictions could slow domestic labs while competitors abroad continue advancing. At the same time, U.S. policymakers have been warning for months about the strategic importance of keeping frontier AI capabilities aligned with national security interests.
Another factor was competitive momentum. Anthropic’s rivals have continued to improve their own cybersecurity-capable models, and in some cases those systems have started to outperform Mythos 5 on security-related evaluations. That created a political and industrial argument for allowing controlled use rather than leaving the model sidelined while others moved ahead.
There was also a direct operational impact on government agencies. According to the reporting surrounding the dispute, top U.S. bodies such as the National Security Agency had lost access to Mythos 5 under the earlier restrictions. For agencies focused on defensive cyber work, losing access to one of the country’s most advanced tools likely added urgency to the push for an exception.
Security tools and strategic pressure
Mythos 5 sits in a category of AI systems designed to help with cybersecurity tasks such as vulnerability analysis, defensive investigation, and infrastructure protection. That makes it particularly sensitive: the same strengths that make the model useful for defenders can also raise fears that bad actors could misuse it.
The administration appears to be trying to thread a narrow needle. It wants to keep the model from broad public exposure while still allowing vetted security organizations to benefit from it. That approach mirrors other recent government decisions in the AI sector, but it also highlights the tension between control and access that has come to define frontier-model policy.
How Anthropic’s case mirrors OpenAI’s latest arrangement
Anthropic is not the only major AI company navigating this new regulatory terrain. Earlier on June 27, OpenAI announced a comparable understanding with the administration involving GPT-5.6, another model with cybersecurity uses. Under that arrangement, access is also limited to approved users in a preview-style rollout.
The parallel is important because it suggests a pattern is emerging: the government may be willing to approve carefully scoped access to frontier systems for selected trusted partners, while still resisting immediate broad deployment. In both cases, the company may move forward in a restricted environment first, then continue pushing for wider availability later.
OpenAI said in its own post that it does not view this kind of government access process as a sustainable long-term model, arguing that it slows the ability of users, developers, enterprises and cybersecurity partners to work with the best tools.
That complaint is likely to resonate across the industry. While many AI leaders have called for regulation, they have also expressed frustration with ad hoc approval systems that can create uncertainty for launches, enterprise contracts and international partnerships.
What the two-week standoff reveals about AI regulation
The Anthropic dispute has become a case study in how quickly AI governance is evolving in the United States. Rather than relying on a single comprehensive rulebook, the Trump administration appears to be using a case-by-case licensing approach for sensitive model releases. That gives regulators flexibility, but it also creates unpredictability for the companies trying to plan product launches and customer deployments.
During the standoff, Anthropic argued for a narrower interpretation of the risk profile around Mythos 5 and Fable 5. The administration, for its part, initially responded with sweeping restrictions. The latest letter suggests both sides have now accepted a compromise: limited access for trusted partners while broader questions remain unresolved.
Even some supporters of stronger AI oversight worry that the current arrangement is too blunt. If the government can cut off access abruptly and then restore it piecemeal, companies may face a difficult environment for long-term investment and product planning. That uncertainty may be one reason why the latest compromise has been welcomed, even by firms that would prefer a lighter-touch model overall.
Why this matters beyond Anthropic
For the broader industry, the Anthropic case could influence how future model releases are handled. If the government settles on a process where frontier systems can be approved for a trusted set of users first, then the same framework could be applied to other models in cybersecurity, defense, and critical infrastructure.
That would effectively create a new category of controlled AI deployment, one that sits between internal research access and public launch. Such a system may satisfy national-security concerns, but it would also raise questions about who qualifies as a trusted partner and how those decisions are made.
It could also deepen the divide between U.S. and international access. If foreign nationals are routinely excluded from early access to frontier models, companies with global workforces may have to build separate workflows, staffing models and compliance systems just to participate in the market.
Fable 5 remains the unresolved issue
While Mythos 5 has been partially restored, Fable 5 remains in a holding pattern. That may matter even more in the long run, because Fable 5 is the model Anthropic would likely use for broader customer adoption and public-facing services. Without clarity on that product, the company still lacks the full commercial path it needs to capitalize on its latest model family.
Anthropic says it wants Fable 5 back in general use, but the government has not yet signaled when — or whether — that will happen. At the moment, the only firm progress concerns limited access to Mythos 5 under controlled conditions.
In practical terms, that means Anthropic has regained part of what it lost, but not the main prize. The company can now move ahead with certain cybersecurity clients and government-related partners, yet its flagship public rollout remains stalled.
Timeline of the Mythos 5 dispute
| Date | Event | What it meant |
|---|---|---|
| June 12 | The Trump administration issues restrictions affecting Mythos 5 and Fable 5 | Access is sharply limited, including for some Anthropic employees who are not U.S. nationals |
| June 12–26 | Anthropic and federal officials negotiate over the restrictions | The company pushes for a narrower, risk-based approach |
| June 26 | Commerce Secretary Howard Lutnick sends a revised letter to Anthropic | The government creates an exception allowing select partners to use Mythos 5 |
| June 27 | The decision becomes public through reporting and company statements | Anthropic begins working to restore access for approved users |
Key details at a glance
- Model restored: Mythos 5, Anthropic’s cybersecurity-focused system
- Still restricted: Fable 5, the public-facing model
- Approval scope: A small group of cyber defenders and infrastructure providers
- Government stance: Broad restrictions remain, but a limited exception is now in place
- Policy direction: The administration reserves the right to change the rules again
What comes next for Anthropic
The next phase will likely focus on implementation. Anthropic now has permission to provision access for the approved users named under the exception, but that still leaves plenty of unanswered questions. Which organizations qualify? How quickly can access be restored? What technical or compliance conditions must be met before the model can be used again?
There is also the bigger issue of whether the government will eventually agree to a broader release for Fable 5. Anthropic is clearly hoping that the same logic that restored Mythos 5 access will later apply to its public model. For now, though, the company is operating under a narrow, government-controlled pathway rather than an open commercial launch.
That makes this less a victory than a managed pause. Anthropic avoided being locked out entirely, but the administration kept enough leverage to shape how and when its most sensitive models can be used. For a company trying to compete at the frontier of AI, that is both relief and constraint at the same time.
And if the last two weeks are any indication, the rules around that balance may continue to shift quickly.
