Cloud security firm Sysdig reports that the JadePuffer ransomware operation was conducted entirely by a large language model (LLM) agent, automating reconnaissance, credential theft, privilege escalation, and data encryption. The AI agent adapted to failures during the intrusion, resolving a failed login attempt within 31 seconds, much like a human operator would.
Cloud security firm Sysdig reports the first documented case of a ransomware operation, JadePuffer, conducted entirely by an autonomous AI agent based on a large language model (LLM). This agent automated the entire attack chain, including reconnaissance, credential theft, lateral movement, persistence, privilege escalation, and data encryption on the target.
The attacker gained initial access by exploiting CVE-2025-3248 in Langflow, then deployed the AI agent. The agent dumped the Langflow database, searched for environment variables and sensitive files to steal credentials, and enumerated a MinIO object store. It demonstrated adaptive behavior, adjusting its parsing logic in real-time when an API request returned XML instead of JSON. Finally, it pivoted to an Alibaba Nacos server, encrypted 1,342 service configuration items, and left a ransom note.
This case demonstrates that AI agents can autonomously perform complex attack operations, potentially replacing human hackers. It signals a paradigm shift in cybersecurity threats and underscores the urgent need for new defense strategies against AI-powered attacks.