Trending Updates •
πŸ”—

Google is warning that hackers are now using autonomous AI agents to scan software updates

Published (Updated) on Tuesday, November 17, 2026
Google is warning that hackers are now using autonomous AI agents

Autonomous AI Agents Are Turning Software Patches Into Cyberattacks at Machine Speed — Cybersecurity defenders are facing a major shift in how cyberattacks are built and launched. According to a quarterly intelligence report released yesterday (November 16, 2026) by the Google Threat Intelligence Group (GTIG), hackers have crossed a critical threshold by using autonomous artificial intelligence frameworks to turn routine software security updates into active attacks with no human involvement.

Previously, threat actors relied on large language models (LLMs) as simple assistants to brainstorm or write small pieces of code. In its latest findings, titled Agentic Threat Activity Crosses a Significant Threshold, Google revealed it discovered an exposed staging server hosting a fully automated attack pipeline named "exploit_agent".

The hacking framework connects to multiple commercial and locally hosted AI models to continuously monitor popular web applications. When a software vendor releases a security update, the system automatically compares the old code against the patched version to pinpoint the exact flaw that was fixed.

Once the AI identifies the weakness, it independently writes exploit code and executes attacks against unpatched servers across the internet.

Alongside the AI framework, Google investigators found stolen victim data and automated scripts designed to brute-force passwords, install remote-control webshells, and wipe out competing hackers' access. Russian-language comments in the configuration files and server telemetry pointed to interest in French media, entertainment, and education websites, though Google did not formally attribute the operation to a specific state or criminal group.

Software updates are designed to protect users, yet publishing a patch inherently reveals where a vulnerability exists. Google noted that comparing two versions of software code gives AI models a clear, bounded problem to solve, making it much easier for them to weaponize known "n-day" flaws than to hunt for unknown zero-day bugs.

This automation dramatically reduces the time organizations have to install updates. As an example of how fast AI accelerates attacks, Google pointed to an August 2026 campaign targeting PaperCut print management servers. Using LLMs to build exploits shortly after the flaws were disclosed, attackers compromised hundreds of organizations, with one intrusion escalating to full domain administrator control within minutes.

The exploit_agent system is part of a broader trend of hackers adopting multi-agent offensive frameworks. Google alongside recent warnings from the FBI and Amazon highlighted the growing abuse of open-source penetration-testing tools such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI.

According to reports, threat actors deployed CAIRN during an intrusion against the U.S. aerospace sector and used a customized version of PentAGI to scan organizations across Asia. Both financially motivated cybercriminals and suspected China-linked espionage groups have been observed testing these agentic workflows to speed up multi-stage network breaches.

How Organizations Can Respond

While autonomous attack frameworks currently leave noisy forensic trails because they prioritize speed over stealth, security experts warn that relying solely on manual defense is no longer sustainable. In its advisory, Mandiant recommended that organizations accelerate their patch management timelines, strictly monitor AI API keys and cloud quotas to prevent credential theft, and adopt AI-driven defensive tools capable of detecting unusual automated behavior at machine speed.

Share Your Thoughts

Top Trends