GTIG says attackers now run multi-agent AI tools in live intrusions
Original title: GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold
CAIRN was seen against US aerospace targets; actors also resell stolen LLM access as LLMJacking.
Google Threat Intelligence Group and Mandiant report that cybercrime, espionage, and influence operators are moving from manual prompting to autonomous multi-agent offensive frameworks. GTIG says it has seen open-source tools CAIRN and PentAGI used in intrusions, plus experimentation with CyberStrikeAI and HexStrike AI, alongside the publicized tool ARTEX. Actors also automate patch diffing to weaponize n-day flaws, steal LLM credentials for resale through underground proxy services, and use models for influence content and supply-chain tampering of coding-assistant configurations.
Unconfirmed: The report does not name specific CVE exploits, victim organizations, or named threat groups in the provided text.
- Named frameworks include ARTEX, CAIRN, CyberStrikeAI, PentAGI, and HexStrike AI.
- CAIRN was observed against targets including the US aerospace sector.
- LLMJacking resells stolen model credentials and cloud capacity via services such as LLM Mixers.
- Frameworks discover low-complexity flaws, generate single-use payloads, and adjust from execution feedback.
- Espionage use includes target research and backdoors via coding-assistant configuration changes.
Impact: Faster multi-stage intrusions and resale of stolen LLM access
- ATT&CK
- T1078 Valid AccountsT1190 Exploit Public-Facing ApplicationT1195.002 Compromise Software Supply ChainT1588.002 Tool
- Vendors
- Organizations
- Google Threat Intelligence GroupMandiant
- Countries
- United States
From the original article
12 min read at cloud.google.com
Written by: Aurora Blum, Kelli Vanderlee, Jess Xia, David Mainor Executive Summary As threat actors increasingly move beyond basic prompting to deploying autonomous multi-agent frameworks, automated vulnerability exploitation pipelines, and alternative monetization schemes like LLMJacking, the adversarial AI landscape has reached a critical inflection point.
By Google Threat Intelligence Group · cloud.google.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.