Security
Google says it saw an AI agent turn security patches into exploits without a human, and the Artex framework spreads among Chinese groups running on DeepSeek
Google says it saw a framework turn patches into exploits with no human in the loop. Artex, a pen-test tool, runs on DeepSeek for Chinese groups.
Two linked claims hold up, with limits. Google’s threat team reportedly says it saw, for the first time, a threat actor use an AI framework to find flaws, write exploits and run them without a person directing each step. And Artex, an open-source AI penetration-testing framework, has been used by at least two China-linked groups, mostly with DeepSeek. We read Google’s finding through a Hackread summary, not the original report, and the actor behind it is not identified.
The essentials
- Google (GTIG): its AI Threat Tracker, published November 16, 2026, reportedly describes a framework called exploit_agent. It compares old and patched code, works out the flaw a fix addresses and tries to build an exploit. Google calls it its first observation of fully automated vulnerability research and exploitation.
- Unknown actor: the operator is not identified. The tooling had Russian-language configuration comments, which do not prove nationality. No success rate was given. A staging server suggested interest in French media, entertainment and education sites.
- Artex: CrowdStrike says the group VAPOR PANDA ran it mainly on DeepSeek plus an unidentified commercial model. BleepingComputer reported attacks on more than 200 organizations in 14 countries. Microsoft separately tied a campaign to Storm-0940.
- DeepSeek: Bloomberg-sourced reporting says hackers favor it for its low cost and relatively lax safeguards. Check Point’s November 5 analysis of 10 exposed Artex consoles found Chinese models dominate, led by the DeepSeek Flash family, with American models on 3.
What is new, and what is not
Reading a patch to find the bug it fixes is an old attacker technique. What Google reportedly flags is the automation of the whole loop, from patch to running exploit, with no human step. In May 2026 Google had already reported a zero-day exploit it believed was built with AI; this report goes further on autonomy.
How to read “common among Chinese groups”
CrowdStrike and Microsoft tie two China-linked actors to Artex, and Check Point saw cybercriminals use it too. That shows the tool is spreading. It does not show it is the standard tool of Chinese groups. The SecurityWeek report cited in the request was not read directly.
What is not established
- Which software or patches exploit_agent targeted: not named.
- The operator’s identity or nationality: not established.
- How often the framework succeeded: not reported.
- Whether the Hackread summary matches Google’s full report: not checked.
- How many Chinese groups use Artex as a standard tool: not established; two are named.
Summary table
| Point | What is stated as of November 17, 2026 | Source |
|---|---|---|
| GTIG tracker date | November 16, 2026 | Hackread |
| Framework | exploit_agent; patch comparison; no human step reported | Hackread |
| Artex campaigns | 200+ organizations in 14 countries | BleepingComputer |
| VAPOR PANDA models | Mainly DeepSeek plus an unidentified commercial model | iTnews |
| Check Point sample | 10 exposed consoles; Chinese models dominate; US models on 3 | Let’s Data Science |
FAQ
Is this the first fully autonomous AI exploit attack?
Google reportedly calls it its first observation of fully automated vulnerability research and exploitation. That is Google’s finding, not proof nobody tried before.
Who is behind it?
Unknown. Russian-language comments in the tooling do not establish the operator’s nationality.
Is Artex a hacking tool?
It is an open-source AI penetration-testing framework for authorized testing. Reports say attackers are using it for real intrusions.
Method and limits
Checked on November 17, 2026. We read Hackread, BleepingComputer, iTnews, Let’s Data Science and Deccan Chronicle through search excerpts and one page fetch. We did not read Google’s report or the SecurityWeek article named in the request; the Google details come from a secondary summary.
Sources
- Hackread: Google Spots AI Agent Turning Security Patches Into Exploits (verified 2026-11-17)
- BleepingComputer: Chinese hackers use ARTEX AI in automated attacks against 200 orgs (verified 2026-11-17)
- iTnews: ARTEX AI hacking framework spreads to another China-linked group (verified 2026-11-17)
- Let’s Data Science: Check Point Details Autonomous ARTEX Attacks by Cybercriminals (verified 2026-11-17)
- Deccan Chronicle: China’s Hackers Use DeepSeek for Attacks, Researchers Say (verified 2026-11-17)
Spotted an error in this article? Report it · Our corrections policy