Researchers have discovered what they call the first agentic ransomware fully managed by a large language model. The malware, named JadePuffer, can search for a target, steal login details, and encrypt files without human help. Unlike older ransomware, it does not just follow fixed commands. Instead, it can decide what to do based on the situation and adjust in real time when it faces obstacles. According to the findings, it entered a system through a zero-day weakness in Langflow, an open-source AI development tool. It then accessed a Nacos server and a MySQL database. In one case, it analyzed a failed backdoor attempt and corrected it in 31 seconds by trying another method.
technologyIntermediate1 min read7/8/2026
Researchers discover autonomous ransomware
JadePuffer is a new ransomware program that uses a large language model to make decisions during an attack.
