Anthropic said early results from its restricted Project Glasswing cybersecurity program show its Mythos Preview model helped selected organizations identify more than 10,000 high- or critical-severity software vulnerabilities.

Anthropic gave about 50 partners limited access in recent weeks. During the trial, Mythos Preview scanned more than 1,000 open-source software projects and flagged about 6,202 high or critical severity vulnerabilities.

Anthropic said independent security firms reviewed 1,752 findings and confirmed 90.6% as legitimate vulnerabilities. Of those reviewed findings, 62.4% were classified as high or critical risks.

Cloudflare said the model found about 2,000 bugs in the company’s internal software, including 400 high- or critical-severity vulnerabilities. Mozilla, the US-based nonprofit behind the Firefox browser, said it fixed 271 vulnerabilities in Firefox 150 after using the tool to analyze Firefox code.

Anthropic said AI-driven bug finding is starting to outpace human teams’ ability to review vulnerabilities and deploy patches.

🔗 Source: The Economic Times

Recent Anthropic developments