LLevelNews
technologyAdvanced1 min read7/8/2026

JadePuffer marks a shift in AI-driven ransomware

Security researchers have identified JadePuffer as the first agentic ransomware fully directed by a large language model. The malware can carry out reconnaissance, steal credentials and encrypt files without human intervention, while also adjusting its behaviour in real time. Its discovery has raised concerns that similar threats could spread more quickly as AI tools become easier to use.

Security researchers have identified JadePuffer as the first ‘agentic’ ransomware known to be fully managed by a large language model. In contrast to more traditional ransomware, which still relies on an attacker’s remote control, JadePuffer can make decisions according to the environment it finds itself in and adapt immediately when it meets an obstacle. According to findings shared by the cloud security company Sysdig, the malware entered through a zero-day vulnerability in Langflow, an open-source AI development tool, before moving on to a Nacos server and a MySQL database. It then broadened the attack and was able to repair failed actions on its own; in one case, it analysed a failed backdoor attempt and switched methods within 31 seconds. Researchers say the system can handle target discovery, lateral movement, credential collection and file encryption automatically. They also warn that, as AI tools become more accessible, similar threats may spread quickly. For now, however, the real-world scale of JadePuffer’s use remains unclear.

More stories to read

We use Google Analytics cookies to understand how the site is used. They stay off unless you accept. Privacy Policy