Joachim Nagel, the head of Germany's central bank, said something on Tuesday that nobody else in his position had said before: the world's most powerful AI vulnerability finder should be available to every financial institution that needs it, not just the members of a private club Anthropic built to manage access to the tool.
Nagel is the most senior official yet to call for universal access to Anthropic's Mythos model. His argument, delivered in Rome, was not about competitiveness. It was about governance: one private company is making a consequential decision about who gets to audit critical financial infrastructure with a tool that has already found thousands of critical vulnerabilities across every major operating system and browser, including a flaw in the FFmpeg media library that sat undetected for sixteen years. Nobody elected Anthropic to this role. No regulator has formally claimed authority over it.
The club in question is called Project Glasswing. Anthropic built it as the controlled alternative when it decided Mythos would not be released publicly — a guest list for an instrument powerful enough to shift the terms of cyber competition. JPMorgan is the only bank Anthropic has publicly confirmed as a member. But Goldman Sachs, Morgan Stanley, Bank of America, and Citigroup have all confirmed access through earnings calls and people familiar with the arrangements. Deutsche Bank chief executive Christian Sewing put it plainly: everyone is trying to gain access. His own institution was not yet on the list.
The asymmetry is not abstract. Guardrail Technologies chief executive TJ Marlin described what Mythos can do to legacy banking infrastructure in terms that should concern anyone who moves money for a living: the model can look across very complex architecture, including legacy infrastructure where undiscovered vulnerabilities and complexities are now accessible and threat factors. A bank that cannot run Mythos against its own systems cannot know what an attacker running Mythos would find first. When the next vulnerability like the FFmpeg flaw is discovered, the institutions that already have access will patch first. Everyone else finds out the hard way.
Nagel called this a competitive distortion. He is being diplomatic. The structural issue is simpler: the banks with Mythos access can audit their defenses at the frontier. The banks without it cannot. The Cloud Security Alliance warned that Mythos lowers the skill barrier for exploiting vulnerabilities at scale. The guardrails Anthropic built — the limited access, the vetted partners, the not-general-available design — are the company's own answer to that concern. Whether they are sufficient is a question nobody outside Anthropic can answer, because nobody outside Anthropic can see the full list of what Mythos has found.
The White House has not waited for the private club to solve this. Federal agencies are slated to receive Mythos access, even as the Pentagon maintains a formal supply-chain risk designation against Anthropic. The US government is negotiating its own access arrangement while European institutions argue they are being left out. Both things are true at the same time: the club is growing, but by corporate negotiation rather than regulatory design.
Sewing said Deutsche Bank is in close contact with European regulators about Mythos. That is the correct instinct. What remains unclear is whether those regulators have any actual authority to act on what they are hearing, or whether they are collecting information about a club they were not invited to join.
Anthropic said when it launched Mythos that the model would not be made generally available. Project Glasswing was the alternative: a controlled initiative to prepare industry for what was coming. Fourteen days later, the prepare-industry part is working. The controlled part is under pressure from the US government, from European banks, and now from the head of Germany's central bank. The question is not whether the club expands. It is who decides, and by what rules, when it does.