Google finds the first fully autonomous AI hacking tool in the wild: it turned fresh security patches into working attacks with no human involved
The 'exploit_agent' framework compared code before and after patches, wrote exploits and ran them on its own.

- Tool name: exploit_agent
- Targets: French websites
- Previous report: 2 months ago
What Google found
Google's Threat Intelligence Group says attackers have crossed a 'significant threshold'. On an exposed server it found a framework called exploit_agent, connected to several AI models, that studied the changes in each security patch, drafted an exploit and launched it automatically, SiliconANGLE reports. The operator appeared to target French media, entertainment and education websites.
Two months ago, Google hadn't seen any fully autonomous attack pipelines.
Why it matters
The report suggests attackers may now prefer recently patched flaws, because AI can turn a patch into an attack within hours. For companies, the gap between a fix being released and installing it is becoming much more dangerous. Google says it has shut the related accounts.






