Anthropic Pushes Claude Mythos Into 150 Critical-Infrastructure Orgs

On June 2, Anthropic announced a major expansion of Project Glasswing — adding roughly 150 organizations across more than 15 countries — pushing its unreleased frontier model Claude Mythos Preview onto the front line of critical-infrastructure defense.

Three key facts

The vulnerability count is staggering. The initial phase (about 50 partners at the April launch) has already surfaced over 10,000 high- or critical-severity flaws spanning every major operating system and browser. Anthropic estimates a major attack on most partners could affect more than 100 million people.

The second wave extends to power, water, healthcare, communications, and hardware — the critical sectors "not well-represented" in round one. Many new members are vendors and nonprofits that maintain code underpinning governments and global critical systems.

Claude Mythos Preview remains unreleased. Anthropic states plainly: "Cheap, fast AI models with powerful cyber capabilities are around the corner." The model is restricted to vetted Glasswing participants, and Anthropic admits it — and, to its knowledge, every other AI lab — has yet to develop safeguards strong enough to keep those capabilities from being misused.

WangDou's Take

Anthropic is holding an AI nuke in one hand and a press release in the other: "Relax, we only handed it to 150 good guys." About as reassuring as your neighbor saying, "I've got an AK, but I'd never sell it to a bad guy." 10,000 vulnerabilities and "could affect 100 million people," written like a doomsday prophecy — plainly a 'bump-your-budget' deck aimed straight at enterprise CIOs.

Source: Anthropic — Expanding Project Glasswing · TechCrunch — Anthropic scales Claude Mythos to critical infrastructure in 15+ countries