Particle.news

Google Warns of Fully Automated Cyberattacks Powered by AI

GTIG says attacker-built frameworks can find patch gaps, write working exploit code and run attacks without human oversight, a change that could let criminals scale rapidly.

Overview

  • Google’s Threat Intelligence Group reported on Tuesday that it has observed the first fully automated exploit campaigns that use AI language models to find and run attacks.
  • The framework called “exploit_agent” automatically compares patched and unpatched versions of common web software, writes exploit code for identified flaws and executes attacks without human intervention.
  • Telemetry from affected systems shows these automated campaigns have stolen data, run automated reconnaissance and brute-force scripts, installed and removed webshells, and focused notably on French media, entertainment and education sites.
  • Google says it suspended accounts and resources tied to misuse of its Gemini models and media reporting says Microsoft has urged much faster patching, while security teams are being told to speed updates, map exposed assets and hunt for signs of compromise.
  • GTIG also found other AI-enabled tools such as CAIRN and signs that state-linked actors are experimenting with automation, which security researchers warn will lower the technical bar and likely let attacks spread and scale worldwide.