GTIG says AI agents wrote and launched exploits with no human
Original title: Google finds AI agent writing and launching exploits with no human in the loop
An exposed server ran exploit_agent against patched web apps; an FBI alert says ARTEX has hit multiple sectors since August 2026.
Google Threat Intelligence Group and Mandiant say threat actors are using agentic pentesting frameworks, including ARTEX, during live intrusions. GTIG describes an exposed staging server running exploit_agent, which compared web-app code around patches, drafted exploits, and ran them, with Russian comments and interest in French media, entertainment, and education sites. The report also covers LLM-quota theft, a suspected North Korean cluster planting SOMBERMEME through Claude Code settings, and automated influence operations.
Unconfirmed: GTIG only suspects a PRC-nexus builder and tracks MIDNIGHT NEPTUNE as a suspected cluster; future stealth gains are anticipatory.
- An FBI FLASH on November 13 said ARTEX use dates to at least August 2026.
- Hundreds of non-working PHP proofs of concept appeared after October 21 patches.
- Some LLMJacking cases passed $100,000 in a day and reached $1 million.
- Stolen model access is resold for $0.10 to $25 per million tokens.
Impact: Automated exploit runs, webshells, stolen data, and burned LLM quotas
- ATT&CK
- T1078 Valid AccountsT1110 Brute ForceT1505.003 Web ShellT1204.002 Malicious FileT1496 Resource Hijacking
- Products
- Claude Code
- Threat actors
- MIDNIGHT NEPTUNE
- Malware
- ARTEXSOMBERMEMEexploit_agent
- Organizations
- Google Threat Intelligence GroupMandiantFBI
- Countries
- ChinaFranceNorth Korea
Vulnerabilities mentioned
All| CVE | Vulnerability | CVSS | EPSS | Signals | Published |
|---|---|---|---|---|---|
CVE-2026-109618+1 related CVE | OOM longjmp memory flaw enabling RCE in PHP | 9.2 | 100% | KEVPoC ×25 | KEV |
From the original article
4 min read at helpnetsecurity.com
Threat actors have started letting AI agents carry out whole stages of live intrusions, and at least one operation no longer needs a human to turn a newly released patch into a working attack, Google Threat Intelligence Group (GTIG) and Mandiant revealed.
By Sinisa Markovic · helpnetsecurity.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.