Skip to main content
Emerging ThreatsMalware & Ransomware

Google Exposes AI-Powered Cyberattacks Leveraging Autonomous Frameworks

Modern tech lab with organized workbench, equipment, and computer terminals.

"Threat actors may find it more efficient to use LLMs to rapidly weaponize patched n-day vulnerabilities, as they can focus research on a finite set of source materials: comparing source code versions (including from pre-release or preview channels), and examining release notes, security research, and POC code related to specific flaws," Google Threat Intelligence Group wrote in a report shared with The Hacker News.

GTIG on agentic exploit frameworks and a shift to "machine speed"

Google's Threat Intelligence Group (GTIG) reports a notable shift: adversaries are moving beyond one-off prompts to deploy autonomous, multi-agent offensive frameworks such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI. GTIG says these agentic frameworks allow attackers to "automate discrete phases and specific operational tasks during live intrusions" and to chain varied attack paths so operations run at what the company calls "machine speed."

Automating patch reversal and the PaperCut NG/MF example

GTIG outlines how threat actors combine large language models (LLMs) and agentic frameworks to automate reverse engineering of patches through differential analysis. That technique lets them rapidly convert newly disclosed, patched vulnerabilities into working exploits. Google points to the PaperCut NG/MF flaws disclosed earlier this September as an observed instance of this behavior, describing LLM-driven pipelines that collapse the time between a patch release and active exploitation.

Open-source tools, ZeroHollow and Agent Zero as enablers

The report highlights how open-source pentesting frameworks and services are being abused as building blocks. GTIG names ZeroHollow, an inference service used for exploit development and LLM routing, and Agent Zero, an open-source agent framework that features autonomous code execution and dynamic tool generation. According to Google, the availability of these frameworks has lowered the barrier to entry and made it easier for "bad actors to easily integrate new exploits, autonomously scale attacks, and mount convincing phishing campaigns."

LLMjacking, underground marketplaces and monetization

GTIG also describes threat actors targeting AI infrastructure directly. The report documents LLMjacking attacks in which attackers steal LLM credentials and cloud resources from compromised enterprise environments. Stolen access is then resold through underground proxy services — GTIG names APIMart and JetNight — sometimes referred to as LLM mixers, creating a secondary market for illicit LLM access and compute.

Jieyou, PhaaS and the rise of low-skill automation

Google singles out a Chinese-language phishing platform called Jieyou, which has integrated an LLM chatbot assistant. GTIG says this evolution permits customers to configure phishing panels and modify templates with natural language commands, enabling low-skilled affiliates to customize pretext lures and phishing interfaces without deep technical expertise. GTIG frames Jieyou as emblematic of a broader trend where phishing-as-a-service (PhaaS) providers treat LLMs as "a modular force multiplier."

What this means for technologists, policymakers, and affected enterprises

  • Technologists and security teams: Watch for dramatically shortened exploit timelines. GTIG's findings underscore that LLM-driven differential analysis can turn patched n-day vulnerabilities into exploitable code quickly, so monitoring disclosures, pre-release channels and associated proof-of-concept (POC) code is more urgent.
  • Policymakers and regulators: The report highlights systemic risks from the proliferation of agentic frameworks, inference services like ZeroHollow, and resale marketplaces such as APIMart and JetNight — all of which change how cyber capabilities scale and are monetized.
  • Affected enterprises and procurement leaders: GTIG's account of LLMjacking and resale of cloud access underlines the need to secure AI credentials and cloud resources, and to scrutinize third-party services that could be exploited as part of an attack chain.

Google's GTIG paints a picture in which familiar cyber tactics are being amplified by agentic automation: the mechanics—reverse engineering patches, exploiting n-day flaws, running phishing campaigns—may not be new, but the speed, scale, and commercial ecosystems around those capabilities are. The report names specific frameworks, services, platforms and marketplaces—ARTEX, CAIRN, CyberStrikeAI, PentAGI, ZeroHollow, Agent Zero, APIMart, JetNight and Jieyou—making clear that the operational landscape has more modular components than before.

The remaining, practical question the facts leave: how rapidly defenders can adapt controls and detection to match attack chains stitched together by autonomous agents and marketplaces that trade in illicit LLM access. For now, GTIG's account is a reminder that the next wave of cyber risk will be measured not only in techniques but in velocity—and in the vendorized tools that make that velocity accessible.

Original report at The Hacker News