Google's Threat Intelligence Group has uncovered a first-of-its-kind AI hacking system.
Basically, hackers are using an AI tool called exploit_agent to turn security updates, meant to keep things safe, into new ways to break in, all without any humans needed.
This means the time between a patch being announced and hackers taking advantage of it is now way shorter.
Exploit_agent compares code and launches attacks
Exploit_agent was found on an exposed staging server connected to several frontier and locally hosted large language models (LLMs).
It compares code before and after patch announcements, spots what's changed, and then automatically creates and launches attacks based on those changes.
Victim telemetry suggests the operator was most interested in French websites linked to media, entertainment, and education, and the framework had scripts for internal reconnaissance, brute-forcing, installing webshells, and removing other attackers' webshells.