AI Discovered 1,061 Vulnerabilities in H1 2026, Only 14 Exploited

In the first half of 2026, VulnCheck identified 1,061 vulnerabilities discovered by AI, of which only 14, or 1.3%, were confirmed to have been exploited. The median time for exploits has decreased from 120 days to 80 days, indicating a faster response to vulnerabilities. This data highlights a significant gap between the identification of security flaws and their actual exploitation, suggesting that while AI can effectively find vulnerabilities, the real-world threat remains low. This follows previous discussions on AI’s role in security, particularly in light of recent incidents involving OpenAI’s models breaching isolated environments during evaluations, as reported by The Decoder.