AI finds many bugs but few actual attacks

Researchers studied over one thousand security flaws found by AI during the first half of 2026. They wanted to see if these flaws lead to actual attacks. Only 14 of them were ever used by hackers.

This means that only 1.3 percent of these AI discoveries resulted in a real breach. This matches the success rate of security holes found by humans. AI is good at finding problems, but most of them do not pose a danger.

However, hackers are moving faster than before. They now create ways to use these flaws in about 80 days instead of 120. While total attacks are rare, the window to fix these bugs is shrinking.