AI Finds Many Security Flaws, Few Exploited: Report

AI is finding many security flaws, but very few of them are actually being exploited. For the first half of 2026, 1,061 vulnerabilities were linked to AI-assisted discovery. However, only 14 of these, or 1.3%, showed confirmed exploitation. This rate is similar to vulnerabilities overall. What's interesting is that attacks are happening faster. Half of all flaws are now exploited within 80 days of disclosure, which is down from 120 days the previous year. About 200 vulnerabilities were attacked within a month. Website content management systems are a major target, accounting for one-third of all cases. AI products themselves, like model-building tools, are also becoming a growing area for attacks. This means the sheer number of findings doesn't always tell defenders about the true risk.