---
id: "CMP-202611-uh4cao"
page: "https://infrastack.ai/live/CMP-202611-uh4cao"
kind: "campaign"
aliases: []
title: "Russian-speaking attacker runs 'exploit_agent', an AI agent that turns new web app security patches into working exploits with no human in the loop, and steals data from French media, entertainment and education websites (Google GTIG)"
severity: "high"
cvss: null
epss: null
exploited: true
status: "open"
category: "infrastructure"
products: []
ecosystems: []
tags: ["campaign", "early-warning"]
not_affected_if: []
published_at: null
first_seen_at: "2026-11-16T14:36:53+00:00"
updated_at: "2026-11-16T14:38:05+00:00"
sources: [{"source": "analyst", "url": "https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926", "first_seen_at": "2026-11-16T14:36:53+00:00", "cites": ["https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926"]}]
exploit_sources: []
---

# Russian-speaking attacker runs 'exploit_agent', an AI agent that turns new web app security patches into working exploits with no human in the loop, and steals data from French media, entertainment and education websites (Google GTIG)

Link for people: https://infrastack.ai/live/CMP-202611-uh4cao (this .md is for agents)

Google Threat Intelligence Group's Q4 2026 AI Threat Tracker (16 Nov 2026) describes exploit_agent, a multi-agent framework that works with several frontier and local LLMs. GTIG found it on an exposed staging server of a suspected Russian-speaking operator. It keeps comparing the code of popular web application software before and after each security update, works out the flaw, writes an exploit and runs it automatically, with no human involved. GTIG calls this the first fully autonomous discovery-to-exploitation it has seen. The server held data stolen from victims and scripts for internal reconnaissance, brute-forcing, installing webshells and removing other attackers' webshells. Victim telemetry points to French websites in media, entertainment and education. GTIG has not publicly named which web applications were targeted, the victims, or the dates. The same report says attackers also use the ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI AI frameworks (PentAGI with proxies against aerospace). The point for defenders: the time between a web app patch and exploitation is getting much shorter.

_Summary, check and fix written by a model from the sources; labelled as such._

## Cause

How the attacker got in: a known vulnerability left unpatched (unpatched-vuln). exploit_agent watches web application security patches and automatically writes and runs exploits for the flaws they fix, and it steals data from French media, entertainment and education websites, but GTIG has not said which applications or victims were hit or how each intrusion began.

How it reached others: it stayed inside (none).

So it does not happen to you: Your board lists the advisories that are in range for your stack, with the fix version.

Last month's incidents with the same cause, and the full checklist: https://infrastack.ai/live/breaches.md#unpatched-vuln

## How to check

- You run internet-facing websites built on popular web application software (CMS, forums, e-commerce, LMS). The specific products are unknown. Higher risk if you are a French media, entertainment or education site.\n- Look for webshells or unexpected PHP/script files in web roots, and for webshells that disappeared (the operator deletes rival webshells).\n- Look for brute-force bursts against admin logins, and for exploit-like requests hitting endpoints changed in a recent security update soon after it came out.\n- Indicators (IPs, domains, hashes): GTIG has not published any in the coverage so far; unknown.

## Fix

- Apply security updates for internet-facing web applications and their plugins within hours to days, not weeks. Assume exploits can exist as soon as the patch is public.\n- Use a WAF or virtual patching for apps you cannot update quickly.\n- Require MFA and lockout on admin logins.\n- Monitor web roots for new files and investigate any webshell. If you find one, treat the site and its stored data as compromised and rotate credentials.

## Timeline

- 2026-11-16T14:36:53+00:00 analyst https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926

## Documents

- [describes] @TheHackersNews: 🛑 Google warns hackers are using autonomous AI agents to turn security patches i https://x.com/TheHackersNews/status/2122330337334464721
- [describes] @BleepinComputer: Google warns that fully-automated AI attacks are coming soon
https://t.co/Rst8y3 https://x.com/BleepinComputer/status/2122328904836829643
- [describes] Google: AI agent turned a security patch into a exploit with no human help https://news.ycombinator.com/item?id=50517797
- [describes] Google warns that fully-automated AI attacks are coming soon https://www.bleepingcomputer.com/news/security/google-warns-that-fully-automated-ai-attacks-are-coming-soon/
- [describes] Google: AI agent turned a security patch into a working exploit with no human help https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926
- [discussion] @GoogleCloudSec: New GTIG AI Threat Tracker:

As 2027 nears, human latency is leaving the attack  https://x.com/GoogleCloudSec/status/2122369113594966244

---

_Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai_
