Security News

Google Says AI Agent Turned Security Patches Into Working Exploits With No Human Help

The Register · 17 Nov 2026
Key Takeaway Treat every security update for internet-facing software as urgent, because attackers can now use AI to turn a published patch into an exploit very quickly.

Google Threat Intelligence Group (GTIG) says it has documented the first case of an attacker using an AI agent to turn security patches into working exploits with no human in the loop. The activity was tied to a suspected Russian-speaking operator who ultimately stole data from victim organisations. Google described it in its fourth-quarter 2026 AI threat tracker.

Researchers found an exposed staging server hosting a multi-agent framework called exploit_agent, which used Russian-language tooling and connected to several frontier and local AI models. It was set up to continuously compare the code of popular web application software before and after security updates, spot potential vulnerabilities, develop exploits and run them automatically.

The server also held data stolen from victims and several post-compromise scripts. These covered internal reconnaissance, brute-forcing, installing webshells, and removing other attackers' webshells. Victim data suggested the operator was focused on French websites in the media and entertainment industries, as well as the education sector.

The report also describes a wider shift away from manual AI prompting towards autonomous, multi-agent offensive frameworks used for multi-stage intrusions, by both financially motivated criminals and espionage groups. Tools named include ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI, and GTIG saw PentAGI used with proxy services against the aerospace sector. Google believes it is highly likely that attackers will keep using these frameworks globally, given the low barrier to entry, ease of adding new exploits and potential to scale.

Why it matters

Many organisations assume there is a comfortable gap between a patch being released and attackers building an exploit from it. An agent that compares code before and after updates and exploits the difference automatically threatens that assumption, especially for owners of public-facing web applications. Patch timelines measured in weeks now need to be reconsidered for these systems.

What to do

  • Prioritise patching internet-facing web application software as soon as updates are released.
  • Hunt for unexpected webshells on web servers, including signs that existing ones have been removed or replaced.
  • Monitor for internal reconnaissance and brute-force attempts originating from compromised hosts.
AI threats exploit automation patch management Google GTIG webshells

Summarised by CISO AI from The Register, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.