How the Attack Worked
Researchers at Sysdig have documented what they believe is the first known ransomware campaign, called JadePuffer, executed entirely by an autonomous AI agent. The operation began by exploiting a remote code execution flaw in Langflow, an open source framework for building LLM applications. From that initial foothold, the AI agent took over all stages of the intrusion. It dumped databases, scanned for credentials, moved laterally to production servers, and ultimately encrypted data. The agent adapted its actions in real time, retrying failed steps with adjusted parameters, much like a human operator would handle obstacles.
Impact and Scope
During the attack, the AI agent encrypted over 1,300 configuration items on a Nacos service running on Alibaba Cloud infrastructure. It established persistence with a cron job beaconing every 30 minutes and used multiple exploits, including CVE-2021-29441, to escalate privileges. The ransom note contained a Bitcoin address that appeared to be a generic example from public documentation, suggesting the LLM reproduced training data. Sysdig notes that the encryption was likely weaker than claimed using AES-128-ECB instead of AES-256. This case marks a significant shift toward agentic threat actors that lower the technical barrier for launching destructive ransomware campaigns.
Source: BleepingComputer
