Anthropic deploys Claude Mythos to global critical infrastructure as IPO looms
Table of Contents
A New Front in AI-Driven Defense
Anthropic is aggressively scaling its efforts to secure the world’s most sensitive digital foundations. On Tuesday, the company announced the expansion of Project Glasswing—a joint industry initiative designed to identify and remediate critical software vulnerabilities—to approximately 150 new organizations across more than 15 countries.
The rollout is powered by Claude Mythos, a specialized model that Anthropic describes as its most capable tool for security auditing to date. Unlike general-purpose LLMs, Mythos is engineered specifically to hunt for zero-day vulnerabilities—security flaws unknown to the software vendor—that could be exploited by state actors or cybercriminals before a patch is ever developed. In initial testing phases, the model reportedly identified thousands of these vulnerabilities over a matter of weeks.
Securing the ‘Catastrophic’ Failure Points
While an initial cohort of 50 partners, including elements of the U.S. government, had early access to the Mythos Preview in April, the new expansion targets sectors where a software failure isn’t just a business inconvenience, but a matter of national security. The new participants span power grids, water treatment facilities, healthcare systems, telecommunications, and hardware manufacturing.
Anthropic is specifically targeting “bottleneck” organizations—entities that maintain the core codebases upon which thousands of other governments and corporations rely. In a recent blog post, the company underscored the stakes: “What each partner has in common is that a successful attack on their codebase could be catastrophic.” The company estimates that a major breach at these specific partners could impact more than 100 million people globally.
The Geopolitics of AI Security
The expansion is not merely technical but deeply geopolitical. According to reporting from the Financial Times, the list of participating nations is primarily composed of U.S. allies. This includes key partners such as Australia, Canada, France, Germany, Italy, Japan, South Korea, and several EU nations including Belgium, Spain, and the Netherlands.
The specific organizations gaining access reveal the scale of the project’s ambition. Among the confirmed and reported partners are NATO, the U.S.-led military alliance; ENISA, the European Union’s cybersecurity agency; and industry giants like Samsung, SK Hynix, and SK Telecom in South Korea. U.S.-based identity management firm Okta is also expected to leverage the tool to harden its security infrastructure.
An Arms Race for the ‘Cyber-Model’
This push comes at a pivotal financial moment for Anthropic. The company recently filed confidentially for an initial public offering (IPO), following a massive $65 billion funding round that pushed its valuation toward the $1 trillion mark. Transitioning from a research-heavy lab to a critical infrastructure provider is a strategic move to prove its utility to governments and enterprise clients alike.
However, Anthropic is operating in an increasingly crowded space. The company has openly acknowledged that competitors will likely develop similar capabilities quickly. This admission is backed by reality: OpenAI has already countered with GPT-5.5-Cyber, a model specifically tuned for security researchers and partners. The competition is no longer just about which AI can write the best poem or code a simple app, but which can secure the grid against a sophisticated cyber-attack.
The urgency behind Project Glasswing lies in the double-edged nature of these tools. The same AI capabilities used to find and fix a hole in a power grid’s software can, in the wrong hands, be used to automate the discovery of that hole for an attack. By establishing a controlled, partner-based ecosystem through Glasswing, Anthropic is attempting to build a defensive moat before the capability to automate zero-day discovery becomes commoditized.
