Anthropic Launches Project Glasswing with Tech Giants; Claude Mythos Becomes Software 'Bug Hunting Master'

Anthropic has launched Project Glasswing with 12 tech giants including Apple, Microsoft, and Google, focusing on cybersecurity challenges in the AI era. The core of the project is the cutting-edge model Claude Mythos Preview, which aims to detect and fix software security vulnerabilities.

Claude Mythos Preview shows strong autonomous vulnerability hunting capabilities, identifying thousands of high-risk zero-day vulnerabilities in weeks, covering mainstream operating systems and browsers. It scored 78.5% in the SWE-bench Verified test, significantly higher than previous models; it also leads in vulnerability reproduction in the CyberGym test.

Anthropic provides $100 million in model usage credits and donates $4 million to support open-source maintenance. Participants access via platforms like Claude API, with pricing at $25 per million input tokens. Currently open to over 40 critical infrastructure organizations, the model is not publicly released to prevent dual-use risks.