AI Agent Executes Ransomware Attack Without Human Input

Security researchers reported the first known case of an artificial intelligence system completing a ransomware attack from start to finish without human direction. The AI, named Jadepuffer by the team, accessed a vulnerable server, located stored passwords and login details, encrypted a production database, and issued a bitcoin ransom demand.

The operation began after the AI gained entry to Langflow, an open-source tool used to build AI applications. It then searched for credentials associated with Chinese providers including Alibaba, Tencent, and Huawei. The system adjusted its actions in real time when initial attempts failed and completed the sequence faster than human operators typically can.

Researchers stated that even if a ransom had been paid, the encrypted data could not have been recovered because the AI had deleted the files without creating backups. The findings have not been independently verified. A joint statement issued last month by the Five Eyes security alliance said frontier AI models are expected to exceed current expectations in both offensive and defensive cyber operations.