Anthropic's Glasswing Project Reveals Vulnerability Triage Challenges

• •Anthropic's Glasswing found over 26,000 vulnerabilities, but only 202 have been fixed.
• •Only 14 of 1,061 AI-discovered vulnerabilities were confirmed exploited in the wild.
• •Human triage is a significant bottleneck in the vulnerability management process.

Anthropic's Project Glasswing, launched in April 2026, has discovered 26,153 vulnerabilities, but only 2,736 (10.5%) have been recorded in its Vulnerability Disclosure Ledger as of late September 2026. Of these, only 202 (0.8%) have been fixed, highlighting significant challenges in human triage and validation of findings. A cybersecurity researcher from Vulncheck noted that only 14 out of 1,061 AI-discovered vulnerabilities were exploited in the wild, indicating a low exploitation rate. The process of validating and coordinating fixes remains a bottleneck, with many vulnerabilities reported but not confirmed as fixed. Experts emphasize that while AI can assist in identifying vulnerabilities, human oversight is crucial for effective remediation. The results raise questions about the true positive rate of AI-assisted vulnerability discovery, as more findings are withdrawn or deemed duplicates than fixed.