Claude Mythos AI Finds 10,000 High-Severity Flaws in Widely Used Software

Anthropic disclosed that Project Glasswing, its AI-led cybersecurity initiative, has identified over 10,000 high- or critical-severity vulnerabilities in globally significant software since its launch last month.
Approximately 6,202 high- or critical-severity vulnerabilities involving over 1,000 open-source projects have been discovered, with 1,726 valid true positives, including 1,094 high- or critical-severity flaws.
Project Glasswing leverages the AI model Claude Mythos Preview to find these vulnerabilities among a partner set of about 50 entities.
Critical flaws, such as the CVE-2026-5194 vulnerability in WolfSSL, have been identified and patched in 97 cases, with advisories issued for 88 vulnerabilities.
Anthropic has emphasized the challenge of fixing vulnerabilities compared to finding them, underscoring the importance of quicker patch cycles to enhance software security.
The AI model Mythos Preview excels in creating end-to-end attack chains and has advanced capability in offensive security tasks, according to autonomous security platform XBOW.
Anthropic highlights the importance of defensive measures and urges software developers to shorten their patch cycles to address security issues more effectively.
Anthropic has started the Cyber Verification Program for security professionals to use its models for legitimate purposes, such as cybersecurity research, but stresses the need for additional safeguards to prevent misuse.

You may have missed