clauding.de
DE
• 2 min read AI-generated

Anthropic Opens Mythos to 150 Organizations — NATO and Samsung Are In

Copy article as Markdown

Project Glasswing is growing fast: Anthropic's most powerful model now scans critical infrastructure across 15 countries. Over 23,000 vulnerabilities found so far.

Anthropic Claude Mythos Glasswing Security Enterprise
Featured image for "Anthropic Opens Mythos to 150 Organizations — NATO and Samsung Are In"

Anthropic announced today that Project Glasswing — the program deploying Claude’s most powerful model to secure critical software — is expanding massively. Around 150 new organizations across more than 15 countries are getting access to Claude Mythos Preview.

From 50 to 200 Partners

When Glasswing launched in April with roughly 50 partners, it was still a contained pilot. Now it’s becoming a global security net. The new partners include Okta, Samsung, the EU cybersecurity agency ENISA, and — yes, really — NATO.

The use cases read like a list of everything you’d never want to see hacked: power grids, water systems, healthcare, telecommunications, and hardware manufacturers.

23,000 Vulnerabilities — and Counting

The numbers from the first phase are striking. Mythos has identified over 23,000 potential security vulnerabilities across partner codebases so far. Anthropic estimates that more than 6,000 of those will be confirmed as severe flaws. That’s a lot of bugs for a model that’s technically still in preview.

Mythos for Everyone — In Weeks

Perhaps the biggest news comes at the end of the announcement: Anthropic says Mythos-class models will be available to all customers “in a matter of weeks.” That would be a massive step — until now, Mythos has been strictly limited to selected partners.

What This Means

Anthropic is increasingly positioning Mythos as a security tool for critical infrastructure, not just a chatbot upgrade. That’s a smart strategy: when you’re scanning NATO codebases, nobody can dismiss you as a toy. At the same time, Anthropic is building a moat that OpenAI and Google can’t easily replicate — because trust with security agencies isn’t won through benchmarks, but through track record.

The interesting question will be how the public Mythos release differs from the Glasswing version. A model that finds vulnerabilities in NATO systems is unlikely to ship unchanged for everyone.

Sources: