Anthropic’s Mythos 5 is back

At a glance:

• Mythos 5 reinstated for a limited set of trusted cyber defenders and infrastructure providers after a June 26 government letter.
• The export control restriction remains, prohibiting foreign nationals—including non‑US employees—from accessing Mythos 5 or Fable 5.
• Both Anthropic and OpenAI are operating under a temporary, case‑by‑case access arrangement pending a broader regulatory framework.

Government Letter and License Revision

A June 26 letter from Commerce Secretary Howard Lutnick to Anthropic co‑founder Tom Brown outlined a revision to the license requirements for Mythos 5 and Fable 5. The government noted that Anthropic had recently collaborated with US authorities to mitigate risks associated with the models. Lutnick wrote that appropriate safeguards were in place to permit certain trusted partners to access the Claude Mythos 5 Model.

He emphasized that all other requirements of the June 12 letter remain in effect until further notice. Lutnick also reserved the right to reevaluate and adjust the scope of license requirements for both Mythos 5 and Fable 5 should circumstances change. This language signals a conditional, case‑by‑case approach rather than a permanent waiver.

Limited Access and Export Controls

The United States did not lift the export control directive that had barred any foreign national from accessing either model two weeks earlier. Instead, an exception was granted that permits a small group of approved organizations and their non‑US national employees to use Mythos 5. This mirrors the limited preview arrangement previously announced for OpenAI’s GPT‑5.6.

Anthropic spokesperson Danielle Ghiglieri said the company had received notice that Mythos 5, its strongest cybersecurity model, could be redeployed to a select set of cyber defenders and infrastructure providers. She added that the firm is actively provisioning the approved participants and restoring their access as quickly as possible. Ghiglieri expressed optimism, stating the company is pleased with the progress and intends to expand access to Mythos 5 and eventually make Fable 5 generally available.

Industry Pressure and Future Outlook

Intense pressure mounted on the Trump administration to modify its case‑by‑case regulatory stance as Anthropic’s competitors improved their cybersecurity benchmarks. Analysts warned that delayed access could allow Chinese AI initiatives to gain ground while top US labs were constrained. National Security Agency officials confirmed they had lost access to Mythos 5, heightening concerns over national security implications.

OpenAI’s recent GPT‑5.6 blog post argued that this short‑term government access process should not become the long‑term default for AI model distribution. The company emphasized that restricting the best tools from developers, cyber defenders, and global partners hinders broader availability. They are pursuing a temporary step they believe will pave the way for a repeatable process and a future cyber Executive Order framework.