Tuesday, November 17, 2026

News

Google: Suspected Russian-Speaking Operator Automated Exploit Attacks With AI Agents

SecurityPatryk Raba
Google: Suspected Russian-Speaking Operator Automated Exploit Attacks With AI Agents
Fot. Asoundd, Wikimedia Commons (CC BY-SA 4.0)

GTIG and Mandiant describe exploit_agent, a framework that compares code before and after patches, writes exploits and runs them without a human. Google calls it the first such case, but that is the analysts' assessment, not an independently confirmed fact.

Contents
  1. How exploit_agent works
  2. What the attribution rests on
  3. A broader trend
  4. What it means for companies

Google Threat Intelligence Group (GTIG) and Mandiant said on November 16, 2026, that they found a tool in which AI agents turn security patches into working exploits on their own and use them immediately. According to Google, a suspected Russian-speaking operator is behind it, and the operation ends with data stolen from victim organizations.

How exploit_agent works

According to The Register, researchers found an exposed server hosting a multi-agent framework for researching and weaponizing vulnerabilities. The tool continuously performs differential analysis of popular web software: it compares code before and after security updates, flags potential flaws, writes exploits and runs them automatically.

The server also held data stolen from victims and post-compromise scripts. According to CyberScoop, they covered internal reconnaissance, password guessing and webshell installation, and one script removed webshells planted by rival attackers.

According to CyberScoop, Google said it had identified threat actors using such tools to fully automate vulnerability discovery and exploitation without a human in the loop for the first time.

What the attribution rests on

The researchers did not name a specific group or state. Linking the operator to Russian-speaking circles rests on Russian-language comments in the code, including a configuration file that tells users to enter API keys for automated exploitation there rather than in a chat. In addition, victim data on the server pointed to French websites.

The Register writes about a suspected Russian-speaking operator. The description comes from Google's report, not from independent verification. The label of a first fully automated attack is likewise an assessment by Google's analysts.

A broader trend

Until now, Google had mainly observed AI used to speed up human work, such as crafting phishing lures, writing code and summarizing stolen data. The company calls the current moment a critical inflection point. The report also lists the frameworks ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI; according to The Register, PentAGI was used against the aerospace sector, and according to CyberScoop, CAIRN and PentAGI appeared in activity targeting a US aerospace organization and organizations in Asia.

Google also says suspected China-linked actors are experimenting with such tools, likely for espionage. As reasons for further spread, GTIG cites the low barrier to entry, the ease of integrating new exploits and massive scaling potential.

What it means for companies

The tool exploits flaws for which a patch already exists, so delays in deploying updates raise the risk of attack. CyberScoop notes that after the PHP patches were published, hundreds of purported exploits appeared, most of which did not work and were likely created with AI help. According to GTIG, cited by CyberScoop, such tools trade stealth for speed and leave noisy forensic traces even when an attack succeeds.

Share: