Initial users of Anthropic’s Mythos continue to retain access following the US directive.

      When Washington instructed Anthropic to disable access to its most advanced AI model, it did not cut off access for everyone. Some organizations that Anthropic had previously selected to test Mythos retained their access to a preview of the system, even as other versions ceased operation under a US export directive, according to a Bloomberg report.

      The details are significant. Anthropic restricted what it calls Mythos Preview to about 200 organizations, including the US government, through its Glasswing program after the model exhibited an unusual ability to identify thousands of software vulnerabilities. A less powerful version of Mythos was distributed more broadly but was later disabled when the Commerce Department mandated that Anthropic suspend access for all foreign nationals due to national security concerns. Yet, the preview access for some users remained unaffected.

      Two specific companies confirmed their continued access. Dragos, the industrial cybersecurity firm that Accenture is in the process of acquiring a majority stake in, and Cisco Systems both informed Bloomberg that they still had access to Mythos Preview.

      This information is indicative: the organizations that maintained access are precisely those utilizing the model for defensive security purposes, which Anthropic emphasizes as the primary justification for why advanced AI in vetted hands is a public benefit rather than a risk.

      The disparity with Europe is stark. The European cybersecurity agency ENISA, which had been invited to participate in Glasswing prior to Washington's blockade, was informed on Friday that it would no longer be granted access. This decision reverses an arrangement that had been the focus of a meeting just days earlier and directly answers the open question from that meeting about whether a European agency could align with a US export order. The current answer is no.

      This incident highlights the significant discretion Anthropic has in the space left by the directive. The export order specifically targets foreign nationals but does not seem to dictate, on an organization-by-organization basis, who among the existing Glasswing members retains access to the preview.

      It was unclear how Anthropic was deciding individual access, which implies that the company is, in effect, making decisions case by case regarding who can utilize one of the most capable security models available.

      This places any vendor in a difficult position and reveals much about their operations. Anthropic has spent the year cautioning about the risks of advanced AI while maintaining that models like Mythos should be accessible to defenders. The directive forced Anthropic to make a choice, and the distribution of access—allowing US security firms to continue while excluding a European agency—reflects the contours of the export rule rather than a judgment of merit.

      The model at the center of this situation is indeed powerful, contributing to the problem at hand. A system capable of identifying thousands of vulnerabilities is equally beneficial to attackers as it is to defenders, and this dual-use aspect is exactly why the Commerce Department moved to impose restrictions. The same capability that prompts Dragos and Cisco to seek continued access is what makes Washington apprehensive about who else might obtain it.

      This divided outcome also raises questions about who controls access to dual-use AI. When a single company determines, in the vacuum created by a government order, which security teams retain access to a frontier model and which do not, the practical authority over a national security-relevant tool lies with a private vendor rather than a public entity.

      That grants considerable discretion to one firm, and the inconsistency between the retained US firms and the excluded European agency highlights this discretion in a way that invites scrutiny.

      Anthropic has not clarified its criteria for access, and the situation continues to evolve as the company navigates the balance between governmental directives and customer needs. What is evident is that the shutdown was not complete, that those who retained access are primarily US defensive security users, and that the first European participant has now been the first to be excluded.