Everyone is still worried that AI will take their jobs—maybe hackers will be the ones “taken” first by AI.

@sysdig The JADEPUFFER being tracked has already upgraded from “using an AI Agent to automatically compromise databases” to deploying ransomware ENCFORGE specifically designed to attack AI infrastructure.
It targets not ordinary Word, Excel, but model weights, training data, vector databases, and checkpoints—covering around 180 file formats. During the first attack, after the agent failed to log in, it fixed its approach by itself within 31 seconds; this time, after downloading the malicious program failed, it consecutively revised 6 versions of scripts in 5 minutes 24 seconds, and finally completed container escape.
But it’s still far from AI waking up by itself. Humans are responsible for choosing the targets, preparing the servers, and initiating the actions—only the technical work after the intrusion is handed over to the AI Agent to carry out.
The company originally worried that its database would be locked; now it also has to prepare offline backups not only for the database, but also for the models it trained over the course of months and spent hundreds of thousands of dollars to build. In the cost breakdown of an AI startup, a new—and very expensive—project suddenly appears.
#Ourbit Not Just Crypto, One-Stop Trading for Global Hot Assets.