Anthropic Wins Partial Reprieve as US Government Greenlights Mythos 5 for Select Users
Table of Contents
A Limited Return for Anthropic’s Cybersecurity Powerhouse
After a tense two-week standoff with the White House, Anthropic has secured a partial victory in its battle against restrictive federal AI controls. The Trump administration has authorized the redeployment of Mythos 5, the company’s high-end cybersecurity model, but only for a narrow group of approved organizations. The move comes after a series of negotiations between Anthropic co-founder Tom Brown and the Department of Commerce.
A letter dated June 26, sent by Commerce Secretary Howard Lutnick to Brown, confirms a “revision to the license requirements” for the model. This shift follows an effort by Anthropic to demonstrate that it has adequately addressed the risks associated with the Mythos and Fable model classes. While this opens the door for a select few, the broader public remains locked out: Fable 5, the public-facing version of the Mythos architecture, is still in regulatory limbo with no confirmed date for a general rollout.
In a statement to the press, Anthropic spokesperson Danielle Ghiglieri confirmed that the government has notified the company that Mythos 5 can be redeployed to a small cohort of “cyber defenders and infrastructure providers.” Ghiglieri expressed optimism regarding the progress and stated that the company is working to restore access to these approved providers as quickly as possible.
The Strategic Pivot: From Blanket Bans to Case-by-Case Licenses
The current arrangement marks a shift in how the administration is handling high-stakes AI models. Two weeks ago, the government hit Anthropic with a sweeping export control directive that effectively barred any foreign national—including the company’s own non-US employees—from accessing either Mythos 5 or Fable 5. This blanket ban created an operational nightmare for a global AI lab.
Now, the government is employing a “trusted partner” framework. By granting exceptions to specific organizations, the administration is mirroring the approach it took earlier today with OpenAI’s GPT-5.6. Under these new terms, non-US nationals employed by Anthropic or the approved organizations are now permitted to interact with Mythos 5, provided they meet the government’s safety criteria.
Secretary Lutnick’s letter emphasizes that this is not a total deregulation. The Commerce Secretary explicitly reserved the right to reevaluate and adjust these license requirements if circumstances change, and noted that the core restrictions from the June 12 directive remain in effect.
Industry Tension and the ‘China Gap’
The decision to loosen the reins on Mythos 5 follows mounting pressure from both the private sector and the national security community. For the past fortnight, key government entities, including the National Security Agency (NSA), have been unable to utilize one of the most capable cybersecurity tools available. This gap created a paradoxical situation where US government agencies were effectively sidelined while foreign adversaries potentially advanced their own AI capabilities.
Furthermore, the restrictive environment began to impact Anthropic’s competitive edge. Industry benchmarks suggested that competitors’ cybersecurity models were beginning to close the gap or even surpass Mythos 5 while Anthropic was bogged down in regulatory disputes.
The trend toward a government-managed “access process” is not being welcomed by the labs. OpenAI, in its announcement of GPT-5.6, explicitly cautioned that such a system should not become the long-term default, arguing that it prevents developers and cyber defenders from accessing the tools they need to stay ahead of threats. For now, however, Anthropic and OpenAI both seem to have accepted this limited-access compromise as the only viable path toward eventually regaining full public availability.
