Anthropic has widened Project Glasswing from its initial 50 participants to 150 organizations across 15 countries. The program gives selected critical-infrastructure operators access to Claude Mythos, the company’s most capable vulnerability-hunting model, to review codebases for zero-day flaws before attackers find them.
The new roster spans Australia, Belgium, Canada, France, Germany, India, Italy, Japan, the Netherlands, New Zealand, South Korea, Spain, Sweden, Switzerland and the United States. Named participants include Okta, Samsung, SK Hynix, SK Telecom, NATO, ENISA and the U.S. government, covering power, water, health care, communications and hardware.
Anthropic says the common thread is the scale of possible harm: for most partners, a major attack on their codebase could affect more than 100 million people. During testing over several weeks, Claude Mythos reportedly found thousands of zero-day vulnerabilities, which explains why the company is limiting access to institutions where defensive use matters most.
The expansion also adds Claude Security, a product layer meant to move from discovery to remediation advice. Anthropic is trying to turn model-led code review into a usable security workflow, while OpenAI pursues its own cyber line with GPT-5.5-Cyber for certified teams.
Sources: CocoLoop, TechCrunch, Yahoo Finance