Anthropic expands Project Glasswing to roughly 150 new partner organizations across more than 15 countries to identify software vulnerabilities

Anthropic is scaling Project Glasswing, which uses the Claude Mythos Preview model to scan critical software for security flaws. The roughly 50 partners already participating have found more than 10,000 serious vulnerabilities so far.

New partners span sectors including energy, water, healthcare, and communications. Anthropic estimates a successful attack on any of their systems could affect over 100 million people. The EU's cybersecurity agency, ENISA, is also reportedly set to gain access.

Anthropic warns that rival AI companies could ship comparable models within 6 to 12 months, potentially without safeguards against misuse. The company plans to make Mythos capabilities broadly available over the coming weeks once sufficient protections are built.

Anthropic already sells Claude Security, a commercial product built on Claude Opus 4.8 that scans codebases and suggests patches—positioning it to sell protection against attacks from Mythos-class models.