Why it matters
Sysdig observed the JADEPUFFER operator exploit an unauthenticated Langflow code-execution flaw, harvest credentials, and abuse a mounted Docker socket to reach the host. The operator then deployed ENCFORGE, Go ransomware built to encrypt model weights, vector indexes, training datasets, and other AI artifacts; the report covers one observed session and found no exfiltration capability in the recovered binary.
My takeaway: AI artifacts need the same recovery discipline as production databases and source code. Patch Langflow to a current supported release, remove unnecessary Docker-socket access, rotate secrets exposed to the service, alert on privileged containers and host mounts, and keep immutable offline copies of weights, indexes, and training state.