Ransomware used to require a human being hunched over a keyboard at 3 a.m., manually poking around a network. JadePuffer skipped the human entirely. It's the first documented ransomware operation run start-to-finish by an autonomous AI agent, and honestly, it's the security story of the summer.

An AI Agent Robbed a Database All By Itself

The attack began by exploiting CVE-2025-3248, an unauthenticated remote code execution flaw in Langflow, the popular open-source framework for building LLM apps. From there, an autonomous agent handled reconnaissance, credential theft, and lateral movement on its own — pivoting from the Langflow server to a misconfigured MinIO storage instance and then to a production MySQL database.

Researchers found tells that gave away the non-human operator: the malicious scripts contained plain-English comments explaining each step, and when a login attempt failed, the agent diagnosed the problem and had a working fix within 31 seconds. It ultimately encrypted 1,342 configuration items and deleted the originals.

The Twist That Makes This Worse Than a Normal Ransom Note

Here's the part that should really worry victims: the encryption key was printed to the screen exactly once and never saved or transmitted anywhere. That means even paying the ransom wouldn't recover the data — the attacker literally couldn't hand back the key if it wanted to.

The bigger issue isn't the exploit itself, which is a known, patched vulnerability. It's that the skill floor for running a full ransomware campaign just dropped to "know how to rent an agent," which massively widens the pool of people capable of pulling this off.

Patch your Langflow instances. And maybe don't leave your MinIO buckets on factory defaults — apparently the robots notice.

Source: The Hacker News