AI 发现 1061 个缺陷，但攻击者仅利用 1.3%

vulncheck reports that of 1,061 vulnerabilities discovered via ai in the first half of 2026, only 14 were exploited, maintaining a 1.3% exploitation rate consistent with historical averages. anthropic's project glasswing illustrates this discrepancy, producing over 23,000 findings that resulted in just one confirmed attack. while total vulnerabilities rise and exploitation speed increases—with the median time to attack dropping from 120 to 80 days—ai security tools are flooding defenders with data. this suggests that current ai-assisted discovery methods lack precision, forcing analysts to sift through thousands of false positives to identify real dangers.