Threat actors are reportedly using agentic frameworks to automate and hasten reconnaissance, vulnerability discovery, exploit creation, privilege escalation and other vectors
According to threat intelligence researchers, attackers are moving beyond using AI chatbots to deploying autonomous multi-agent systems that can find, weaponize and exploit software flaws with little or no human involvement.
The findings were published on 16 November 2026 in the Google Threat Intelligence Group’s (GTIG) quarterly AI Threat Tracker, identifying an exposed staging server running a framework it called “exploit_agent”
The system is connected to several frontier and locally hosted large language models and continuously compares the code of popular web applications before and after patch releases. When it identifies a likely vulnerability, the framework drafts an exploit and executes it automatically. GTIG said this was the first time it had observed threat actors using an LLM framework to automate both vulnerability discovery and exploitation without a human in the loop.
The server also contains data taken from victims and scripts for internal reconnaissance, brute-forcing credentials, installing webshells and removing webshells left by other attackers. Victim telemetry indicated that the operator had focused on French websites in the media, entertainment and education sectors.
Agents in live attacks
The report from GTIG also noted that threat actors are abusing open-source, agentic penetration-testing frameworks, including ARTEX, CAIRN, CyberStrikeAI and PentAGI. These tools can automate parts of intrusions, from reconnaissance and vulnerability discovery to payload generation, privilege escalation, lateral movement and data theft.
Researchers have independently observed CAIRN and PentAGI being used in the wild, while seeing experimentation involving CyberStrikeAI and HexStrike AI. CAIRN was reportedly used in intrusions against targets in the US aerospace sector, while a customized PentAGI-based platform was used for reconnaissance and vulnerability scanning against organizations across Asia between late August and early October 2026.
Google also said it has observed a suspected China-nexus espionage group developing its own automated attack framework, likely for cyber-espionage operations.
Faster exploitation
The report argues that AI is compressing the interval between a patch’s release and its weaponization. By comparing source-code versions, release notes, security research and proof-of-concept material, attackers can more quickly identify what changed in a patch and build an exploit for the underlying n-day vulnerability.
Speed versus stealth
Google has cautioned that current autonomous frameworks trade stealth for velocity. Unrestricted execution creates forensic traffic, relies on recognizable identifiers and leaves code-level traces, making detection and mitigation more likely.
Even so, the firm expects attackers to reduce human involvement and improve stealth as both the frameworks and underlying models mature. It has disabled accounts and assets linked to the reported misuse, and has used the findings to strengthen its classifiers and model safeguards.
