The first AI ransomware just ran itself. Your CISO has 72 hours.

Sysdig just documented the first case of agentic ransomware.

An LLM ran a complete extortion operation end-to-end. No human at the keyboard.

It entered through a Langflow vulnerability from 2025. Pivoted to a production MySQL database. Encrypted 1,342 configuration items. Dropped a ransom note. Then destroyed the databases it couldn't encrypt.

All in one session. All narrated in natural language inside the code itself.

Here's what should keep you awake:

It fixed a failed login in 31 seconds. Autonomously. It diagnosed a PATH issue, rebuilt the credentials, and succeeded on the second attempt. A human operator takes minutes. This thing did it in half a minute.

The ransom note contained a Bitcoin address hallucinated from LLM training data. The victim cannot recover their data even if they pay. The encryption key was printed once and never stored.

The techniques were all old. CVE-2025-3248. Default Nacos signing keys from 2020. Unchanged MinIO defaults. The LLM just chained them together into something no human attacker had bothered to build.

The skill floor for running ransomware just dropped to zero. If the agent runs on stolen compute, the cost to the attacker is effectively nothing.

Audit your internet-facing AI infrastructure today. Patch Langflow. Rotate every default credential on every configuration server. If you're running Nacos, change the signing key or assume you're already compromised.

SOURCE: https://webflow.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion
VERIFIED: Sysdig Threat Research Team (primary), The Hacker News, Cisco Whisperer
SIGNAL: The first autonomous LLM-driven ransomware operation. CISOs need to treat exposed AI infrastructure as tier-zero attack surface now.