Developing Cybersecurity

Google says AI agents exploited patches

Google's GTIG and Mandiant said threat actors used AI frameworks to automate vulnerability discovery and exploitation without a human in the loop for the first time. An exposed staging server ran exploit_agent and held stolen data and scripts.

Latest update: Google says AI agents ran exploits with no human in the loop

Image from Illustration: Google says AI agents ran exploits with no human in the loopIllustration

The story so far

On 16 November 2026, Google said a suspected Russian-speaking operator used an AI agent to compare code before and after security updates, build working exploits and steal data from victims without a human in the loop. An exposed staging server held stolen data and scripts, with telemetry pointing to French media, entertainment and education websites. Attackers were also using ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI to speed patch exploitation.

On 17 November 2026, Google's GTIG and Mandiant said threat actors used AI frameworks to automate vulnerability discovery and exploitation without a human in the loop for the first time. They said an exposed staging server ran exploit_agent, which compared application code before and after patches, drafted exploits and ran them. The FBI warned that ARTEX had been used since at least August 2026 against financial services, government, technology, defense, healthcare and energy.

Latest updates

2 updates · newest first
  1. 17 Nov
    New development

    Google says AI agents ran exploits with no human in the loop

    Google's GTIG and Mandiant said threat actors used AI frameworks to automate vulnerability discovery and exploitation without a human in the loop for the first time. An exposed staging server ran exploit_agent, which compared application code before and after patches, drafted exploits and ran them. The FBI warned that ARTEX had been used since at least August 2026 against financial services, government, technology, defense, healthcare and energy.

    Help Net Security ↗CSO Online ↗Infosecurity Magazine ↗
  2. 16 Nov
    New development

    Google says AI agent built exploits from patches with no human help

    Google said a suspected Russian-speaking operator used an AI agent to compare code before and after security updates, build working exploits and steal data from victims without a human in the loop. The exposed staging server held stolen data and scripts, with telemetry pointing to French media, entertainment and education websites. Attackers are also using ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI to speed patch exploitation.

    The Hacker News ↗The Register ↗BleepingComputer ↗

Key facts

Google said a suspected Russian-speaking operator used an AI agent to build exploits without a human in the loop.Reported by The Hacker News, The Register and BleepingComputer
An exposed staging server held stolen data and scripts, with telemetry pointing to French media, entertainment and education websites.Reported by The Hacker News, The Register and BleepingComputer
Attackers were also using ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI to speed patch exploitation.Reported by The Hacker News, The Register and BleepingComputer
GTIG and Mandiant said threat actors used AI frameworks to automate vulnerability discovery and exploitation for the first time.Reported by Help Net Security, CSO Online and Infosecurity Magazine
An exposed staging server ran exploit_agent, which compared application code before and after patches, drafted exploits and ran them.Reported by Help Net Security, CSO Online and Infosecurity Magazine
FBI warned ARTEX had been used since at least August 2026 against financial services, government, technology, defense, healthcare and energy.Reported by Help Net Security, CSO Online and Infosecurity Magazine

What’s still unknown

Whether the suspected Russian-speaking operator is linked to the FBI warning about ARTEX
What data was stolen from the French media, entertainment and education websites
How the exposed staging server was discovered

All sources

6 articles

In the daily editions

Related developing stories

Follow this story by email

Get one email when there’s a new update, no more than one a day.