U.S. officials ordered Anthropic to restrict broad access to its powerful Mythos AI model, but a small group still has access to it. The situation shows how national security rules and cutting‑edge AI now pull in different directions.
How Mythos Ended Up Behind A Wall
Anthropic built the Claude Mythos Preview as a frontier model for tough cybersecurity and code analysis work. According to Amazon Web Services, the tool is part of a “gated research preview” that focuses on defensive security, autonomous coding, and long‑running agents. The company said it designed Mythos for “ambitious projects” where finding serious software flaws quickly really matters.
Project Glasswing is Anthropic’s program that gives select partners access to this restricted model. TechCrunch reported that early Glasswing users included firms such as Amazon, Apple, Cisco, CrowdStrike, Microsoft, and Palo Alto Networks, as well as the Linux Foundation. These partners agreed to use Mythos for defensive work and to share what they learn so the rest of the industry can improve its security tools.
However, Reuters and Bloomberg later reported that U.S. officials grew worried about how far Mythos and its sibling models could go. As a result, Washington issued an order that forced Anthropic to shut down a broader, less restricted Mythos variant for foreign users under new export control rules. Those rules aim to prevent attackers from using advanced systems that can find vulnerabilities to target U.S. critical infrastructure.
Project Glasswing Keeps A Narrow Channel Open
Even after that move, about 200 Glasswing organizations still have access to Mythos Preview, according to a Bloomberg report cited by Reuters. Those users include major security companies such as Dragos and Cisco Systems, both of which confirmed they remain connected to the model. At the same time, the European Union’s cybersecurity agency, ENISA, said it was told that its planned access would be revoked following the U.S. order.
Anthropic’s own Mythos page explains that Mythos 5 is “only available to a small group of vetted partners” because it is highly capable in cybersecurity and biology, and “could be used both for good and for harm.” The company adds that it is working to “safely open up access” over time through trusted programs rather than a public release.
Yet sources told Reuters that it remains unclear exactly how Anthropic is determining which Glasswing members retain access under the new government limits.
READ MORE: Pi Network Price Has Stalled: June 28 Will be Crucial
