For the first time, Google’s threat researchers say they have seen attackers run an AI system that turns newly published security fixes into working attacks with no person in the loop. In a report published on 16 November, the Google Threat Intelligence Group described a tool called exploit_agent, found on an exposed staging server and connected to several AI models. It compared versions of popular web software before and after each patch was announced, worked out the likely flaw, wrote an exploit and ran it automatically. Victim data suggests the operator focused on French media, entertainment and education websites. In its previous quarterly report, the same team said it had not yet seen fully autonomous attack pipelines in use.
The report also describes a suspected North Korean group that altered Claude Code settings files inside code repositories, so that the coding tool downloaded a backdoor each time it started.
the time between a fix being published and attackers using the flaw is shrinking, so putting off updates is riskier than it used to be. Organisations should patch internet-facing web software quickly and treat AI coding-tool configuration files as security-sensitive. Read Google’s report.



