Google's cybersecurity division has observed for the first time that attackers have fully automated the discovery and exploitation of security flaws using AI language models. That is the finding of the latest report from the Google Threat Intelligence Group (GTIG), which describes a framework called exploit_agent.
How the framework works
According to GTIG, exploit_agent automatically compares unpatched and patched versions of widely used internet software to find flaws. When it finds one, the program writes attack code on its own and runs it, with no human stepping in. In the attacks identified, data was pulled from the targeted sites.
More than the initial break-in
GTIG also found several scripts for actions meant to run automatically after a successful intrusion. The report mentions internal reconnaissance, brute-force attacks, installing web shells and removing the web shells of other attackers. The victims' collected telemetry suggests the operator mainly targeted French websites in the media and entertainment industry and in education.
Not an isolated case
According to GTIG, other cybercriminals are also trying to automate attacks with AI, for example using freely available pentesting tools. State-backed actors from North Korea misuse generative AI, among other things to scale their operations. Google expects automation to keep growing. The reasons: the barrier to entry is low, new exploits are easy to plug in, and the potential to scale is enormous. GTIG therefore considers it very likely that attackers will keep using such frameworks in campaigns against organizations worldwide.
For you, this mainly means one thing: if attackers evaluate patches automatically, the window between an update and an attack gets smaller. So install security updates promptly.