Google Caught an AI Agent Turning Security Patches Into Exploits With No Human Help. Patch Faster.
Updated 7 AM PT (10 AM ET), Monday, Nov. 16, 2026. Google’s threat researchers say they caught something new: an AI framework that watches for software patches, works out the flaw each one fixes, writes an exploit and runs it, with no person steering. Google’s Threat Intelligence Group (GTIG) calls it the first time it has seen attackers fully automate vulnerability discovery and exploitation “without a human in the loop,” in its AI Threat Tracker report. The defensive answer is old and boring: patch faster.
What Google found
- The tool: A framework called exploit_agent, found on an exposed staging server. Per GTIG it “conducts continuous differential analysis of popular web application software,” compares code before and after patch announcements, drafts exploits and “automatically executes them without human intervention.” It connects to multiple frontier and locally hosted LLMs. The report does not name them.
- The targets: The operator was “specifically interested in targeting French websites associated with the media and entertainment industries, as well as the education sector.” The same server held scripts for internal reconnaissance, brute-forcing, installing webshells and removing other attackers’ webshells.
- A WordPress example: One figure in the report shows the agent analyzing an access-control check in the Spectra (Ultimate Addons for Gutenberg) WordPress plugin, version 2.19.25. The report does not say that attempt succeeded.
The bigger picture in the report
GTIG says AI-driven vulnerability research is shrinking the gap between a patch and an exploit. Zero-day exploits rose modestly, from an average of 8 a month in 2025 to 11 a month in 2026, with a spike to 22 in August. Overall disclosures and exploits nearly doubled between January and August 2026. Two examples: researchers and security firm Huntress produced working exploits for a pair of PaperCut zero-days within a day of disclosure, and GreyNoise reported a campaign four days after disclosure in which an attacker “likely used LLMs” against hundreds of organizations and reached domain administrator “within minutes.” Per SiliconANGLE, GTIG’s previous quarterly report, from Sept. 8, said it had not yet seen fully autonomous attack pipelines in the wild.
What isn’t known
Google gives no success rate for exploit_agent and attributes it to no group. A Russian-language comment in its configuration file tells the operator to enter keys “right here, not in the chat,” but the report attributes exploit_agent to no one. The report also says other actors, including financially motivated groups and suspected China-linked espionage actors, are building and testing their own agentic frameworks. Google says it disabled the accounts and assets tied to the activity it found and used the findings to strengthen its model classifiers. SiliconANGLE notes that the noisy trail left by unrestricted autonomous attacks may make them easier to detect for now.
PCGeek’s take
This answers a question readers keep asking after the Microsoft, Workday and FAA outages: are hackers using AI? Here is a documented case of attackers using it, and it is a different thing from those outages. Officials found no evidence of AI or a hack behind the FAA ground stop, and Workday blamed its own backup test. The practical lesson is the one Google gives defenders: put patched but unapplied flaws at the top of the list, because the window to fix them is getting shorter. If you run a WordPress site or any web app, turn on updates for core and plugins, and check that they are actually installing. Google also flags AI tool configuration files, such as .claude/settings.json, as a place attackers have tampered; keep an eye on changes to them in shared repositories.
Cover photo: Austin McKinley, CC BY 3.0, via Wikimedia Commons. File photo of Google’s Mountain View campus, not the systems involved in the report.
How quickly do you apply updates on your sites and devices, and would an AI that exploits patches within hours change that? Tell us in the comments. Follow @pcgeekinfo for updates.