Google Uncovers AI Agents Automating Exploit Development from Patches and LLMjacking Operations

Google Threat Intelligence Group has detailed how threat actors are moving beyond simple AI prompts to deploy autonomous multi-agent frameworks such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI for faster multi-stage intrusions. Adversaries are now using large language models to perform differential analysis on software patches, enabling rapid weaponization of disclosed vulnerabilities including the PaperCut NG/MF flaws observed in September. Open-source pentesting tools and services like ZeroHollow and Agent Zero are being repurposed to automate reverse engineering, vulnerability analysis, and proof-of-concept creation at machine speed. Nation-state actors aligned with China and financially motivated groups have been observed experimenting with these frameworks to scale operations and chain multiple attack paths. In parallel, attackers are conducting LLMjacking campaigns to steal cloud credentials and resell access through underground services such as APIMart and JetNight. The Chinese-language Jieyou phishing platform has integrated an LLM chatbot to let low-skilled users customize campaigns via natural language commands.

The Hacker NewsTranslated from English

Read at source

Google has revealed that adversarial use of artificial intelligence has advanced from manual prompting to fully autonomous multi-agent offensive frameworks. The company’s Threat Intelligence Group documented threat actors deploying tools such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI to conduct multi-stage intrusions at significantly higher operational velocity.

Attackers are also leveraging large language models to automate reverse engineering of software patches through differential analysis. This technique allows rapid weaponization of newly disclosed vulnerabilities, as demonstrated with the PaperCut NG/MF flaws disclosed earlier in September. Google noted that actors can focus on a limited set of source materials including version comparisons, release notes, and existing proof-of-concept code.

The abuse of open-source agentic penetration-testing frameworks is viewed as a paradigm shift that enables attacks to occur at machine speed. Google observed both financially motivated actors and suspected Chinese espionage groups experimenting with these tools to scale operations and combine varied attack paths.

Additional tools mentioned include ZeroHollow, an inference service for exploit development, and Agent Zero, an open-source framework supporting autonomous code execution. Threat actors are simultaneously targeting AI infrastructure itself through LLMjacking attacks that steal credentials and resell access via underground proxy services such as APIMart and JetNight.

A Chinese-language phishing platform called Jieyou has integrated an LLM chatbot assistant, allowing customers to configure panels and modify templates using natural language without deep technical expertise. Google described this as part of a broader trend where phishing-as-a-service providers treat LLMs as modular force multipliers.

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in AI & Model Risk

Researchers Show Malicious Webpage Could Hijack Meta Muse AI Agent and Extract Private Conversations
China-Linked Actors Use Autonomous AI Agents to Discover Zero-Day Path Traversal in MONAI Label and Achieve Root Access
AI & Model Risk

China-Linked Actors Use Autonomous AI Agents to Discover Zero-Day Path Traversal in MONAI Label and Achieve Root Access

Suspected China-linked hackers deployed autonomous AI agents powered by the open-source Cairn framework to identify and exploit a previously unknown path traversal vulnerability in MONAI Label, an open-source medical imaging server used by research teams. The flaw, tracked as CVE-2026-133968 and rated 9.8, affects versions 0.8.5 and earlier by allowing unauthenticated attackers to write files outside the designated storage directory when authentication is disabled by default. Amazon Threat Intelligence reconstructed the campaign after discovering an exposed Cairn control panel that revealed the agents' prompts, task settings, and execution history spanning roughly 48 hours in September. Within about 11 minutes the agents mapped the application and achieved arbitrary file writes; in a follow-up session they planted a malicious Python startup file that was loaded during a training job, granting root access on a university server. Maintainers merged a source-code fix on October 19 that enforces path resolution inside approved folders, yet no patched release has been published and the latest available version remains 0.8.5 from 2024. Amazon attributes the activity to a China-linked operator with moderate confidence based on Simplified Chinese prompts and China Standard Time settings, though no specific group or government sponsorship was confirmed.

The Hacker NewsNov 21, 2026Translated from English

Chinese-Linked Actors Use Cairn AI Platform to Discover and Exploit MONAI Label Zero-Day