Content on Harava News is AI-generated and may contain errors.

Google Finds Attackers Using AI Framework to Build and Run Exploits

TechnologyAI

Google says it found attackers using an AI framework to study software patches, find flaws, create ways to exploit them and run those attacks without human help, in the first fully automated flaw-finding and exploitation activity its threat researchers have observed.

In a report published November 16, Google’s Threat Intelligence Group said it found the system, called “exploit_agent,” on an exposed test server. The framework was set up to compare software code before and after patches, search for flaws shown by the changes, write code to use those flaws and run it automatically. Google said the system combined several large language models, including commercial models and models run locally.

The server also held data taken from targeted organizations and scripts for tasks such as searching inside their computer systems, guessing passwords and installing webshells. Webshells are tools attackers can use to keep access to systems they have broken into. Google said clues on the server pointed to an operator interested in French media, entertainment and education websites.

But it did not name a specific group or publicly identify a particular flaw that the framework had successfully exploited. Google’s report also described other tools that act as agents and were used in intrusions, including activity targeting a U.S. aerospace organization and organizations in Asia. The company warned that automation can reduce the time defenders have to apply patches.

It also noted that these tools can leave signs that defenders can detect. The findings mark a significant increase in the use of AI in cyberattacks. However, public evidence does not establish that exploit_agent itself caused any specific breach.