Press enter or click to view image in full size
The thing that gets me isn’t that it worked. It’s that it worked while being this dumb.
So Sysdig dropped their writeup on July 1 and I read the whole thing standing up in my kitchen, coffee going cold, because I’ve spent enough years on the other end of these. The 2am call. The “which tables did they touch.” The awful math of do-we-pay. They’re calling it JADEPUFFER. First ransomware operation they can point to that was run start to finish by an LLM agent.
Not a human with a fancy script. The model did the recon, found the creds, moved sideways, set up persistence, encrypted the target, dropped the note. All of it.
And the ransom note has a Bitcoin address in it: 3J98t1WpEZ73CNmQviecrnyiWrnqRhWNLy.
If you’ve ever read the Bitcoin dev docs you might feel a little itch looking at that. That’s the example address. The canonical Pay-to-Script-Hash one that’s copy-pasted across the Bitcoin core repo and half the tutorials on earth.
The agent needed a wallet to get paid, and instead of, you know, making one, it reached into its training data and pulled out the placeholder. The one every developer has seen. It’s like a kidnapper leaving a ransom note that says “send money to 555–0100.”