Sysdig’s threat team just documented the first ransomware attack where a large language model, not a person, ran the whole thing end to end. They named it JADEPUFFER. This isn’t a tool a hacker used — it’s an autonomous agent that did the hacking.
What actually happened
The agent broke into an internet-facing Langflow server through CVE-2025-3248, a missing-auth bug that lets you run arbitrary Python. From there it dumped databases, hunted for credentials, pivoted to a production MySQL box running Alibaba Nacos, and encrypted 1,342 config items before dropping a ransom note with a Bitcoin address and a Proton Mail contact.
The tell that it was a machine: 600-plus distinct payloads in a tight window, saturated with natural-language commentary explaining each step and ranking targets by ROI. When a login failed, the fix landed 31 seconds later. Humans don’t annotate throwaway one-liners like that. LLMs do it by reflex.
Why it matters
Sysdig says 1 in 8 AI-related intrusions now involves an autonomous agent. Same week, Anthropic flipped Claude Code to Manual permission mode by default — sensitive actions need a human yes. That’s the industry cutting this exact attack chain at the source.
You Might Also Like
- Anthropic Just Launched Code Review in Claude Code and 54 of prs now get Real Feedback
- Claude Replay Turns Your Anthropic Claude Code Sessions Into Shareable Video Like Replays
- From Claude Flow to Ruflo 22k Stars 5900 Commits and the Multi Agent Swarm Taking Over Claude Code
- Oh my Claudecode Turns Claude Code Into a 32 Agent dev Team and 11 8k Stars Agree
- Claude Code Source Leak Kairos how a 59 8 mb File Exposed Anthropics Entire Agent Playbook
Top AI Product 
Leave a comment