---
title: Google finds an agent framework that turns patches into exploits without a human
description: >-
  Google found a multi-agent framework on an exposed staging server that compares web application
  code before and after a security update, then builds and runs...
url: https://expiredqueues.com/security-advisories/gtig-autonomous-exploit-agent-2026-11/
markdown: https://expiredqueues.com/security-advisories/gtig-autonomous-exploit-agent-2026-11/index.md
image: >-
  https://expiredqueues.com/images/og/security-advisories-gtig-autonomous-exploit-agent-2026-11-01c4ad09.jpg
---

[Security advisories](https://expiredqueues.com/security-advisories/index.md)

# Google finds an agent framework that turns patches into exploits without a human

First listed 17 November 2026 Updated 17 November 2026

Closed

Google found a multi-agent framework on an exposed staging server that compares web application code before and after a security update, then builds and runs an exploit by itself. Google suspects a Russian-speaking operator and calls it a first for fully autonomous exploitation. The targets appeared to be French media, entertainment and education sites.

- Vendor

  Google Threat Intelligence Group

- Type

  Trend

- Status

  Closed. No exploitation reported.

- Tags

  ai agents

- History

  - 17 November 2026 New. From Google's fourth-quarter AI threat tracker, as reported by The Register.

[Read the source](https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926)

This is one entry in a list I pick by hand. It tells you whether a fix exists. For how to apply it, follow the source. [See all advisories](https://expiredqueues.com/security-advisories/index.md).

**Follow Expired Queues.** We are on X and Instagram.

[](https://x.com/expiredqueues "Expired Queues on X (@expiredqueues)")[](https://www.instagram.com/expiredqueues/ "Expired Queues on Instagram (@expiredqueues)")

## Structured data

```json
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Google finds an agent framework that turns patches into exploits without a human",
  "description": "Google found a multi-agent framework on an exposed staging server that compares web application code before and after a security update, then builds and runs an exploit by itself. Google suspects a Russian-speaking operator and calls it a first for fully autonomous exploitation. The targets appeared to be French media, entertainment and education sites.",
  "datePublished": "2026-11-17",
  "dateModified": "2026-11-17",
  "author": {
    "@type": "Person",
    "name": "Adrian Noel",
    "url": "https://expiredqueues.com/about/"
  },
  "publisher": {
    "@type": "Person",
    "name": "Adrian Noel"
  },
  "mainEntityOfPage": "https://expiredqueues.com/security-advisories/gtig-autonomous-exploit-agent-2026-11/",
  "isPartOf": {
    "@type": "CollectionPage",
    "@id": "https://expiredqueues.com/security-advisories/"
  },
  "about": "Google Threat Intelligence Group",
  "citation": "https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926"
}
```
