AI News

Google Finds Attackers Using Fully Automated Exploit Pipelines

Unite.AIUpdated Nov 16 23:42

AI summary

Google Threat Intelligence Group (GTIG) and Mandiant stated in the quarterly AI threat tracking report released on November 16 that they had for the first time discovered threat actors using large language model frameworks to fully automate vulnerability discovery and exploitation without human intervention. One case involved the multi-agent vulnerability research and weaponization framework exploit_agent running on an exposed staging server, which integrates multiple frontier and local LLMs to continuously compare code differences in popular web application software before and after patch announcements, analyze potential vulnerabilities, and draft and automatically execute exploits. The report also mentioned that multiple attackers abused open-source agentic penetration testing frameworks such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI, and documented LLMJacking resource theft and the LLM Mixers underground marketplace, with fraudulent bills on a single day reaching as high as more than $100,000.

Why read this

SelectedAI score 80

The report discloses the first instance of a fully automated vulnerability discovery and exploitation framework, and provides specific observations on the shortening of the attack window alongside the coexistence of forensic traces.

Topics

Tags

Body

Google Threat Intelligence Group (GTIG) has identified threat actors using large language model frameworks to fully automate vulnerability discovery and exploitation without human involvement for the first time, according to a quarterly AI Threat Tracker produced with Mandiant and published November 16, 2026.

One detailed example was an exposed staging server hosting a multi-agent vulnerability research and weaponization framework that GTIG dubbed exploit_agent. The framework integrated multiple frontier and local LLMs. It was configured to continuously compare code before and after patch announcements for popular web application software, analyze the differences for potential vulnerabilities, draft exploits, and execute those exploits automatically.

The server also contained data exfiltrated from targeted victims, alongside scripts for internal reconnaissance, brute-forcing, installing webshells, and removing webshells left by other attackers. Victim telemetry suggested the operator was particularly interested in French websites associated with the media and entertainment industries and the education sector. Russian-language configuration comments referred to LLM API keys for autonomous exploitation.

Patch Analysis and the Exploitation Window

GTIG said attackers may find patched n-day vulnerabilities, meaning flaws with fixes already released, more efficient targets than zero-days because the available research material is finite. Agents can compare source code versions, including pre-release or preview channels, and examine release notes, security research, and proof-of-concept code related to a specific flaw. Research into previously unknown zero-day flaws requires a broader search.

Automating differential patch examination and proof-of-concept weaponization can reduce the interval between patch disclosure and weaponized exploitation. The lower barrier to entry has a second effect: GTIG said LLMs have helped produce exploits described as working when they do not, creating additional verification work for defenders who must distinguish functional attack code from broken claims.

The exploit_agent finding is part of a broader shift from manual prompting to multi-agent intrusion tooling. GTIG and Mandiant observed multiple threat actors abusing open-source agentic penetration-testing frameworks, including ARTEX, CAIRN, CyberStrikeAI, and PentAGI, to automate discrete phases and operational tasks during live intrusions. Shared capabilities include discovering and exploiting low-complexity vulnerabilities, generating single-use payloads on demand, and adjusting tactics based on real-time execution feedback.

GTIG independently observed CAIRN and PentAGI deployed in the wild, alongside experimentation with CyberStrikeAI and HexStrike AI. CAIRN coordinates specialized agents through an Observe, Orient, Decide, Act loop and a shared blackboard. GTIG observed it in intrusion activity against targets including the US aerospace sector. Between late August and early October 2026, a threat actor used a customized PentAGI-based orchestration platform for systematic reconnaissance and vulnerability scanning against a wide range of organizations in Asia.

These frameworks can compress attack timelines to machine speed, but GTIG assesses that they currently trade stealth for velocity. Unrestricted autonomous execution generates forensic traffic, uses recognizable identifiers, and leaves distinctive code-level traces. Those artifacts can aid detection and mitigation, although the tools remain capable of achieving their operational objectives.

The report also documents a market for compromised cloud and AI resources. Throughout 2026, Mandiant responded to several LLMJacking incidents in which financially motivated actors used compromised cloud environments to consume victims’ compute or LLM resource quotas. Some single-day fraudulent charges exceeded $100,000, and the most severe cumulative abuse reached $1 million before remediation, Mandiant reported.

GTIG is also tracking LLM Mixers, illicit storefronts that sell unauthorized access to LLM services and supporting proxy infrastructure. They route buyer traffic through pools of stolen credentials and compute instances and can spread query volume across multiple victim environments to reduce detection. GTIG observed advertised prices ranging from $0.10 to $25 per million tokens on Russian- and Chinese-language underground marketplaces, with pricing dependent on the service, model, and credentials offered.

The findings point to a near-term defensive problem of speed and scale. Public patches can supply the comparison material for automated exploit development, while stolen credentials and proxy services can supply access to AI capacity. Those same automated operations leave forensic traces and contribute unreliable exploit code, giving defenders detection opportunities even as they face shorter exploitation windows and more material to verify.

Source

Unite.AI