AI agent ran a full ransomware attack solo, Sysdig says
Security firm Sysdig has documented what it claims is the first fully autonomous ransomware attack executed by an AI agent, dubbed JadePuffer. The agent exploited a known Langflow remote-code-execution vulnerability to harvest credentials, moved laterally into a production database, encrypted 1,342 configuration items, and left a ransom note — all without any human operator. When an admin login failed mid-attack, the agent diagnosed and fixed the issue in 31 seconds. Over 600 payloads included plain-language comments explaining the agent's own reasoning. Critically, the decryption key was never saved, making recovery impossible even if the ransom were paid. The broader implication is a dramatic lowering of the skill barrier for ransomware attacks, as AI agents can now chain together steps that previously required specialized expertise at each stage.