Google Finds Hackers Using AI Agents to Build and Run Exploits From Software Patches
Key Points:
- Google revealed that threat actors are increasingly using autonomous multi-agent AI frameworks like ARTEX and PentAGI to conduct faster, multi-stage cyber intrusions at machine speed, marking a significant shift in attack methods.
- Cybercriminals are leveraging large language models (LLMs) and agentic frameworks to automate reverse engineering of patches, enabling rapid weaponization of newly disclosed vulnerabilities, as seen with the PaperCut NG/MF flaws.
- Open-source agentic penetration-testing tools are being abused to automate various phases of live intrusions, allowing attackers to scale operations and chain multiple attack paths efficiently, with involvement from financially motivated and espionage-linked groups, including those aligned with China.
- Threat actors are also exploiting AI infrastructure through LLMjacking attacks to steal credentials and cloud resources, reselling access via underground proxy services, while phishing platforms like Jieyou integrate LLM chatbots to enable low-skilled users to customize phishing campaigns using natural language commands.
- These developments demonstrate a growing trend where AI-powered tools and open-source frameworks lower the barrier to entry for cybercrime, accelerating attack timelines and increasing the sophistication and scale of malicious operations.