Google Finds AI Tool That Autonomously Turns Software Patches Into Exploits
November 16, 2026
Google’s Threat Intelligence Group (GTIG) says it has observed attackers using exploit_agent, an AI framework that autonomously analyzes software changes after security patches, identifies likely vulnerabilities, drafts exploits and executes them.
Google described this as its first observation of threat actors fully automating vulnerability research and exploitation in real-world activity, while noting that this reflects its investigations rather than proving no earlier attempts existed.
The approach focuses on known but unpatched “n-day” vulnerabilities: patch releases give attackers code changes, release notes and research to analyze, making this route potentially more efficient than hunting for unknown zero-days.
The report illustrates how quickly this can unfold: researchers generated working exploits for two PaperCut vulnerabilities within a day of disclosure, and a suspected AI-assisted campaign reported four days later targeted hundreds of organizations; one intrusion reached domain-administrator access within minutes.
Google found exploit_agent on an exposed staging server that also held stolen victim data and scripts for reconnaissance, password brute-forcing and webshell activity; material on the server suggested interest in French media, entertainment and education websites, but Google did not confirm that every generated exploit worked.
This marks a shift from GTIG’s previous quarterly report, which had not identified fully autonomous attack pipelines against real-world targets, and the group expects attackers to keep turning to patched flaws that organizations have not yet fixed.
Google also observed other offensive tools, including ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI, in activity involving cybercriminals and suspected espionage actors; reported operations included CAIRN use against targets that included the US aerospace sector and PentAGI activity across Asia.
Financially motivated criminals and suspected espionage groups, including groups with a PRC nexus, have experimented with these frameworks, which are easy to adopt and scale; GTIG warned their low barrier to entry makes continued use against organizations worldwide highly likely.
Still, faster execution does not guarantee a successful or stealthy attack: autonomous tools can leave conspicuous forensic traces, and their results may be unreliable.
AI is also making phishing easier to customize: the Chinese-language phishing-as-a-service platform Jieyou added a chatbot that lets customers build phishing panels and lures with natural-language commands.
Google described AI-supported influence operations, including a platform developed with Gemini API support that was intended to manage more than 100,000 social-media accounts; that figure was planned capacity, not a confirmed active-account count.
Google said it disabled accounts linked to misuse of its services and strengthened safeguards, while urging organizations to install updates promptly, restrict access when updates must be delayed, and investigate possible prior compromise because patching alone does not remove an attacker already inside.
Summary based on 8 sources
Get a daily email with more Tech stories
Sources

theregister • Nov 16, 2026
Google: AI agent turned a security patch into a working exploit with no human help
SiliconANGLE • Nov 16, 2026
Google says an AI framework turned security patches into exploits with no human in the loop
India Today • Nov 17, 2026
Google says hackers are using AI agents to weaponise security patches without human intervention