Anthropic's secret model backs cyber coalition

# Anthropic’s secret model backs cyber coalition Anthropic has introduced a new artificial intelligence system called Claude Mythos Preview, but unlike a normal model launch, it is not being released to the public. Instead, the company is putting it inside Project Glasswing, a new cybersecurity effort that includes major technology firms such as Amazon Web Services, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, Palo Alto Networks, JPMorganChase, and the Linux Foundation, along with more than 40 additional organizations that maintain critical software infrastructure. Anthropic says the model is strong enough at finding and exploiting software flaws that broad release would be too risky. (anthropic.com) That is a sharp break from the usual artificial intelligence playbook. Most leading model companies compete by putting their newest systems in front of as many users as possible, then building products and subscriptions on top. Anthropic is taking the opposite route here: keeping its most capable model gated and using it first for a narrow, high-stakes job—defensive cybersecurity. (anthropic.com) Project Glasswing is built around a simple idea: if artificial intelligence can now find software weaknesses faster than humans can, defenders need access to those tools before attackers do. Anthropic says Mythos Preview has already found thousands of high-severity vulnerabilities, including flaws in every major operating system and web browser. In a separate technical write-up, the company says the model was able to identify and exploit zero-day vulnerabilities, meaning previously unknown flaws, and that some of the bugs it found were decades old, including a now-patched 27-year-old bug in OpenBSD. (anthropic.com) Those claims explain why Anthropic is describing Mythos as a cybersecurity turning point rather than a consumer product. In its system card, the company calls Claude Mythos Preview its most capable frontier model to date and says the jump in capability was large enough that it decided not to make it generally available. Instead, Anthropic says it will use the model with a limited set of partners while it studies safeguards and deployment practices. (anthropic.com) Anthropic is also putting real money behind the effort. The company says it is committing up to $100 million in usage credits for Mythos Preview across Project Glasswing and $4 million in direct donations to open-source security organizations. That funding matters because much of the internet’s critical infrastructure depends on open-source software maintained by small teams with limited budgets, even when the software itself is used by banks, cloud platforms, hospitals, and governments. (anthropic.com) The coalition structure is important too. Anthropic is not pitching Mythos as a tool for one company’s internal security team alone. It is framing the model as shared defensive infrastructure for the firms that build and maintain the software stack everyone else depends on. The launch partners include cloud providers, chip companies, security vendors, software platforms, and open-source stewards, which suggests Anthropic wants Glasswing to sit upstream of ordinary enterprise security products, closer to the code and systems that power the internet itself. (anthropic.com) That positioning says a lot about where the artificial intelligence market is heading. For the last two years, the public face of large language models has been chatbots: writing emails, summarizing documents, generating images, and answering questions. But the more lucrative and defensible business may be specialized systems used in environments where mistakes carry legal, operational, or national-security consequences. Anthropic’s move places it less in the category of chatbot vendor and more in the category of high-assurance infrastructure provider. That is an inference from the way it is restricting access, choosing partners, and emphasizing controlled deployment over mass adoption. (anthropic.com) The word “assurance” matters here. Big companies buying artificial intelligence for security, finance, healthcare, or government work do not just want the strongest model on a benchmark. They want auditability, access controls, safety processes, and confidence that a tool will not create a bigger risk than the one it is supposed to solve. Anthropic’s own materials tie Mythos Preview to its Responsible Scaling Policy and Frontier Compliance Framework, which shows the company is trying to package capability with governance rather than capability alone. (anthropic.com) There is also a competitive angle. Anthropic has long tried to distinguish itself from rivals by emphasizing safety and controlled deployment. CNBC noted that the company spent years building a reputation for responsible release practices, and this announcement arrived only weeks after a public clash involving the United States Department of War and Anthropic’s safety posture. In that context, Glasswing is not just a product decision; it is also a brand statement about what kind of artificial intelligence company Anthropic wants to be. (cnbc.com) The timing is notable because Mythos had already surfaced in public through leaks before Anthropic formally announced it. Fortune reported in late March that an unsecured data trove exposed details of an unreleased Anthropic model called Mythos, and Anthropic later acknowledged it was testing a model that represented a “step change” in capabilities. The official launch of Project Glasswing turns what began as a leak-driven mystery into a controlled narrative: yes, the model is real, yes, it is unusually capable, and no, it is not going into a public chatbot. (fortune.com) There is still plenty Anthropic is not disclosing. The company says more than 99 percent of the vulnerabilities Mythos found have not yet been patched, so it is withholding technical details under coordinated disclosure practices. That means outside researchers cannot yet independently verify the full scale of the claims. For now, the strongest evidence comes from Anthropic’s own documentation and the fact that major partners were willing to join the program at launch. (red.anthropic.com) Even with that caveat, the direction is clear. Anthropic is betting that the next important artificial intelligence market will not be the broadest one, but the most trusted one: tools used quietly inside banks, cloud platforms, browsers, operating systems, and critical infrastructure. If Claude Mythos Preview performs as Anthropic says it does, Project Glasswing may end up being remembered less as a product launch than as a signal that frontier artificial intelligence is moving from public demos into the machinery that keeps digital systems standing. (anthropic.com)