---
title: Google GTIG: attackers now run agentic frameworks in live intrusions, and an exposed 'exploit_agent' server showed patch-diff-to-exploit with no human in the loop | ctxsec
url: https://ctxsec.com/items/gtig-ai-threat-tracker-nov-2026-exploit-agent-autonomous-patch-diff-exploitation
description: "GTIG's latest AI Threat Tracker says adversaries have moved from prompting to autonomous multi-agent offensive frameworks (it names ARTEX, CAIRN, CyberStrikeAI and PentAGI) in live intrusions. GTIG found an exposed staging server hosting 'exploit_agent', a multi-agent framework wired to frontier and local LLMs that diffs popular web apps before and after patches, infers the bug, drafts an exploit and runs it automatically; the server also held stolen victim data and post-compromise scripts. GTIG's statistics show zero-days exploited rose from a 2025 average of 8 a month to 11 a month in 2026, spiking to 22 in August. The report also covers LLMjacking resale, PRC-nexus actors using LLM platform storage for staging and exfiltration, and an actor manipulating coding-assistant configurations to drop backdoors."
status: "research (Google Threat Intelligence Group / Mandiant report, Nov 16, 2026); attributions are GTIG's"
published: 2026-11-16
---

> GTIG's latest AI Threat Tracker says adversaries have moved from prompting to autonomous multi-agent offensive frameworks (it names ARTEX, CAIRN, CyberStrikeAI and PentAGI) in live intrusions. GTIG found an exposed staging server hosting 'exploit_agent', a multi-agent framework wired to frontier and local LLMs that diffs popular web apps before and after patches, infers the bug, drafts an exploit and runs it automatically; the server also held stolen victim data and post-compromise scripts. GTIG's statistics show zero-days exploited rose from a 2025 average of 8 a month to 11 a month in 2026, spiking to 22 in August. The report also covers LLMjacking resale, PRC-nexus actors using LLM platform storage for staging and exfiltration, and an actor manipulating coding-assistant configurations to drop backdoors.

[← AI](/ai) · [markdown](/items/gtig-ai-threat-tracker-nov-2026-exploit-agent-autonomous-patch-diff-exploitation.md)

research (Google Threat Intelligence Group / Mandiant report, Nov 16, 2026); attributions are GTIG's

2026-11-16 · published here

2026-11-19 11:43 AM ET

### Google GTIG: attackers now run agentic frameworks in live intrusions, and an exposed 'exploit_agent' server showed patch-diff-to-exploit with no human in the loop

- **Takeaway:** the time between a public patch and a working exploit for internet-facing web apps is shrinking. Patch edge and web apps in days, not on the monthly cycle.
- **Note:** some X summaries say Russian operators used 22 zero-days; GTIG's 22 is the overall August zero-day count, not exploit_agent's tally.

Sources: [Google Threat Intelligence Group: GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold (Nov 16, 2026)](https://cloud.google.com/blog/topics/threat-intelligence/agentic-threat-activity-crosses-a-significant-threshold)
