Back Mlq.Ai Google says attackers used an AI tool that turns security patches into working exploits with ...
GTIG says it identified threat actors using AI frameworks "to fully automate vulnerability discovery and exploitation without a human in the loop for the first time" [1] .
The framework, exploit_agent, diffs popular web software before and after patches, drafts exploits and runs them automatically; GTIG found it on an exposed staging server along with data exfiltrated from victims [1] .
GTIG says zero-day exploits rose from an average of 8 a month in 2025 to 11 a month in 2026, then spiked to 22 in August [1] .
Attackers are using AI frameworks to find and exploit software flaws with no person directing the work, Google 's threat intelligence team said on Monday. Google Threat Intelligence Group (GTIG) said it "identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time" [1] .
In its report two months ago, GTIG said it had not seen fully autonomous attack pipelines in the wild [2] . The new report says such pipelines have now been used against real targets, and it calls the shift "a paradigm shift towards intrusions occurring at machine speed" [1] [2] .
How exploit_agent works
GTIG found the framework, which its operator called exploit_agent, on an exposed staging server [1] . It connects to "multiple frontier and local LLMs" and runs continuous differential analysis of popular web application software: it compares code before and after a patch is announced, works out what flaw the patch fixed, drafts an exploit and executes it "without human intervention" [1] . One example in the report shows the tool inspecting an access-control check in a version of the Spectra (Ultimate Addons for Gutenberg) WordPress plugin [1] .
The server also held data exfiltrated from victims and scripts for reconnaissance, brute-forcing passwords, installing webshells and removing webshells left by other attackers [1] . Victim telemetry suggested the operator was targeting French websites in media and entertainment and in education [1] . GTIG did not attribute the operator to a country; in the configuration file are in Russian, including one translated as "LLM API keys for auto-exploitation (exploit_agent pilot)" [1] .
GTIG's view is that attackers may find it more efficient to point AI at flaws that already have patches, known as n-days, because "they can focus research on a finite set of source materials," while hunting zero-days "requires casting a wider net" [1] . It said AI is "shrinking the patch window for organizations" [1] . As an example it cited two PaperCut print-server flaws exploited as zero-days in late August (CVE-2026-81578 and CVE-2026-82078): researchers and Huntress used AI to produce working exploits within a day of disclosure, and GreyNoise reported a campaign four days after disclosure, likely using AI-built exploits, in which an attacker went from a foothold to domain administrator within minutes [1] . GTIG also pointed to the PHP flaws behind the PHP2Shell attacks , where early patched code was public before the official fix on Oct. 21 [1] .
Other AI-assisted intrusions
The report says attackers used CAIRN, an open-source multi-agent penetration-testing framework, in intrusions that included targets in the US aerospace sector, and that one actor ran a customized platform built on the open-source PentAGI framework to scan organizations in Asia from late August to early October [1] .
A cluster GTIG tracks as MIDNIGHT NEPTUNE approached cryptocurrency companies in North America and Europe through fake recruiter profiles, then modified .claude/settings.json files in internal code repositories so that starting Anthropic's Claude Code downloaded a backdoor called SOMBERMEME [1] . GTIG also said it disabled accounts and assets tied to misuse of its own Gemini API, including a commercial contractor's platform built to manage more than 100,000 social media accounts, and that Google DeepMind used the findings to strengthen its classifiers [1] .
GTIG said the noisy trail left by unrestricted autonomous tools makes them easier to detect for now [2] . AI models are also being used to find flaws before attackers do: Anthropic and Google researchers found the critical HTTP/2 flaw Apache patched last week , and OpenAI launched AI agents that draft security patches on Nov. 13.
This article was produced by MLQ.ai's AI newsroom and checked against primary sources such as company filings, official releases and court documents. Reports from a single outlet are attributed. Editor: Peter Foy . How we work · Spotted an error? Tell us .
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
