The U.S. government has given Anthropic permission to bring back its most powerful AI system, Claude Mythos 5, in a tightly controlled, limited re‑release, restoring access for about 100 vetted organizations after a sweeping export‑control order forced the company to shut the model down just two weeks earlier. Mythos 5 will now return primarily as a cybersecurity and critical‑infrastructure tool for selected agencies and companies, under new safeguards that highlight how Washington is trying to manage the risks of frontier AI without permanently sidelining cutting‑edge systems.
From abrupt shutdown to controlled comeback
Anthropic’s agreement with Washington caps a tumultuous two weeks. On June 11, the company announced it was “suspending access to Claude Fable 5 and Claude Mythos 5” after the U.S. government issued an export‑control directive that barred “all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.” To comply, Anthropic disabled both models for all customers worldwide, even American ones, effectively pulling its most advanced systems off the market days after launch.
Mythos 5 had only just been introduced in April to a limited audience of companies and researchers, with Anthropic touting it as particularly adept at detecting software vulnerabilities and supporting cybersecurity teams. Three days after a broader June launch, regulators ordered it offline amid national‑security concerns that adversaries such as China or Russia could weaponize its capabilities to find or exploit weaknesses in U.S. systems.
On Friday, according to The New York Times, Anthropic and the Trump administration finalized a deal that partially reverses that shutdown. The Commerce Department sent the San Francisco–based firm a letter granting permission to “restore access to its Mythos 5 model for certain clients,” while discussions continue over the status of Fable 5 and potential broader access in the future.
Commerce Secretary Howard Lutnick wrote that Anthropic had “collaborated with the U.S. government to mitigate risks associated with the covered models,” adding that “these efforts have produced substantial progress.” In return for the green light, Anthropic pledged to work with officials on guidelines for future AI model releases.
Who gets Mythos 5 now, and who doesn’t
The new permission is far from a general reopening. NBC News reports that “Mythos 5 access will be restored to around 100 organizations that include government agencies and private companies,” citing people familiar with the arrangement. Reuters and Semafor say those include “more than 100 U.S. institutions,” such as major companies and government agencies, and confirm that many are involved in cybersecurity and critical‑infrastructure defense.
Devdiscourse and NewsBytes describe the approved organizations as “trusted partners” and “approved cyber defense and infrastructure groups,” noting that the list includes several Fortune 500 firms but excludes ordinary commercial customers. Access remains blocked for most foreign organizations and for regular users of Anthropic’s consumer products; Mythos 5 is not returning as a wide consumer rollout.
In a statement posted on X and quoted by TechCrunch and KuCoin, Anthropic acknowledged the limited scope of the re‑release. “Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5,” the company wrote. “Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure. We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”
Importantly, TechCrunch notes that the administration’s directive “did not address the release of Fable 5,” meaning Anthropic’s other top‑tier model remains fully suspended for now. Foreign nationals outside the approved institutions also remain barred from using Mythos 5 under the revised controls.
New safeguards: how Mythos 5 is being constrained
While the government has not published a full list of conditions, officials and reports point to several safeguards underpinning the re‑release.
The Commerce Department’s letter, described by the Times and Devdiscourse, says Anthropic “implemented safety measures on its AI” in exchange for restored access. CNN reports that regulators “revised license terms” to narrow who can deploy Mythos 5 and under what circumstances, ensuring “strict safety standards” for approved users.
KuCoin and NewsBytes say the lifting of restrictions followed Anthropic’s work with regulators on “risk‑on assets assessments” and technical mitigations to reduce the chance that the model could be used for offensive cyber operations or support banned activities. Those steps likely include tighter controls on vulnerability‑related outputs, enhanced monitoring for misuse patterns, and contractual limits on how the model can be integrated into tools.
Anthropic itself has emphasized collaboration. In its June 11 suspension statement, the company said it was “complying with the government’s legal directive” and “working to restore access as quickly as possible in a manner consistent with national security.” Its new X post signals continued willingness to align release practices with government expectations, even at the cost of limiting commercial reach.
For users inside the approved organizations, Mythos 5 will still appear as a powerful tool for vulnerability detection, threat‑modeling and defensive cyber operations, but under the shadow of monitoring and potential constraints that are unusual for a commercial AI service.
A clash de‑escalated, and a precedent set
The limited re‑release de‑escalates what the New York Times calls a “clash between the Trump administration and the company over its cutting‑edge artificial intelligence systems.” Anthropic had been forced to disable Mythos 5 and Fable 5 “within a mere 90 minutes” of receiving the export‑control order, an abrupt move that shocked customers and raised fears about heavy‑handed AI regulation.
By negotiating a controlled comeback, both sides avoid a prolonged standoff. The Commerce Department can claim it has reduced risks while allowing advanced AI to support U.S. cyber defenses; Anthropic can say it has preserved at least some access to its flagship model and established a channel for future approvals.
But the episode also sets a precedent. As Wired and CNN note, the Mythos 5 deal shows that frontier AI releases may now be subject to bespoke negotiations with Washington, particularly when models are seen as dual‑use technologies with serious national‑security implications. It dovetails with the administration’s broader push to have leading AI firms “voluntarily submit new models” for review, and with its recent request that OpenAI stagger the release of GPT‑5.6.
For Anthropic, the green light is both relief and constraint. The company regains the ability to serve high‑value cybersecurity customers with Mythos 5, but only under a regime that keeps the model out of general circulation and ties its access to continuing government confidence. For the wider AI industry, it is a signal that the days of unilaterally pushing frontier systems to anyone who can pay may be over, replaced by a future in which the most powerful models are introduced, paused and, re‑introduced on terms shaped as much by regulators as by engineers.