TechTarget reports that five months into Anthropic’s Project Glasswing, security researchers are finding a wide gap between the volume of AI-discovered vulnerabilities and the number that are actually exploited or patched. VulnCheck’s Patrick Garrity found that of 1,061 vulnerabilities attributed to AI-assisted discovery, only about 14 were listed as exploited in the wild as of June 2026, and that fewer than 1% of the Mythos findings in Anthropic’s disclosure ledger were marked fixed, pointing to human triage rather than exploitation as the bottleneck. Security experts quoted in the piece warn that using more AI to help with triage risks expanding the attack surface.
Glasswing results put AI 'vulnpocalypse' to the test
TechTarget reports that five months into Anthropic's Project Glasswing, security researchers are finding a wide gap between the volume of AI-discovered vulnerabilities and the number that are actually exploited or patched. VulnCheck's Patrick Garrity found that of 1,061 vulnerabilities attributed to AI-assisted discovery, only about 14 were listed as exploited in the wild as of June 2026, and that fewer than 1% of the Mythos findings in Anthropic's disclosure ledger were marked fixed, pointing to human triage rather than exploitation as the bottleneck. Security experts quoted in the piece warn that using more AI to help with triage risks expanding the attack surface.
Source: Techtarget