Patching Alone No Longer Enough as AI Shrinks the Window for Attack
Organisations "can't patch their way out of AI", according to Dick O'Brien, principal intelligence analyst at Symantec. Speaking at the IRISSCON 2026 conference in Dublin on November 11, he said the window for patching has closed rapidly over the past year because attackers are using AI to find zero day flaws. He also said advanced AI models can chain together several low severity vulnerabilities to find an exploit path that humans would struggle to spot.
The Zero Day Clock, a live dashboard tracking the time from vulnerability disclosure to confirmed exploitation, estimates that the average time fell from one week in 2025 to one day in 2026. It projects that this will drop to a minute in 2027. The PHP2Shell vulnerability, disclosed in October 2026, showed this speed: the first reported exploitation attempts came just 19 minutes after it was made public. Amazon also reported detecting an attempted exploitation four days before the vulnerabilities were published.
O'Brien expects attackers to move towards AI agents that carry out attacks independently once inside a network, making them harder to detect and respond to. In a September 2026 attack on the Dutch Institute of Vulnerability Disclosure, AI agents performed post-compromise tasks at machine speed. Symantec's analysis of the institute's report found it took 19 minutes to detect the attack, 19 hours before the response began, and a further nine hours to contain it.