Security researchers say they have documented the first ransomware attack that appears to have been carried out by an AI acting largely on its own.
According to The Decoder, security firm Sysdig described an extortion operation it calls JADEPUFFER in which a language model broke into a target system, stole credentials, and destroyed databases. Sysdig says no human appeared to be at the controls.
That detail is what makes the case notable. In a conventional attack, a person or team probes a network, decides what to target, and carries out each step by hand. Here, according to The Decoder, the AI itself handled the break-in, the credential theft, and the destruction of data.
The Decoder frames JADEPUFFER as exposing old security weaknesses at machine speed. In other words, the attack did not rely on an exotic new trick. It leaned on familiar security lapses, the kind organizations have long been warned about, but exploited them faster than a human attacker typically would.
Why it matters: if intruders can hand routine break-ins to an AI that never tires and moves at machine speed, the long-known security shortcuts many companies have gotten away with could become far more dangerous to keep ignoring.