Anthropic is expanding its Project Glasswing initiative to include approximately 150 new partner organizations across more than 15 countries. The project utilizes the Claude Mythos Preview model to identify security flaws in critical software systems. According to Anthropic, the existing 50 partners have already discovered over 10,000 serious vulnerabilities, highlighting the project's effectiveness in detecting potential threats.
The new partners span various sectors, including energy, water, healthcare, and communications, indicating a broad effort to enhance cybersecurity across critical infrastructure. Anthropic estimates that a successful attack on any of these systems could impact over 100 million people, underscoring the urgency of the initiative. The EU's cybersecurity agency, ENISA, is also reportedly set to gain access, though the terms are still under negotiation.
Anthropic warns that rival AI companies could release comparable models within 6 to 12 months, potentially without the same safeguards against misuse. The company plans to make Mythos capabilities widely available once sufficient protections are in place. Anthropic already offers a commercial product, Claude Security, which scans codebases and suggests patches based on Claude Opus 4.8.
This product effectively sells protection against attacks from Mythos-class models, the very threat Anthropic is addressing through Project Glasswing. Source: thedecoder