What Anthropic Glasswing reveals about the future of vulnerability discovery

Anthropic has launched Project Glasswing, an initiative using Claude Mythos Preview (an AI model designed for cybersecurity) to automatically discover software vulnerabilities at scale, which it is testing with a closed group of over 40 companies including Amazon, Microsoft, and Google. Early testing claims the model found thousands of high-severity vulnerabilities in widely-used software, including some that had been missed for decades, suggesting that AI-powered vulnerability discovery may shift how security work is organized and force organizations to focus less on managing backlogs and more on reducing the time vulnerabilities remain exposed before being fixed. The initiative raises questions about the future role of human-driven security work as AI automation becomes more capable.