When AI Infrastructure Becomes the Target: Securing Gateways and Control Points
Microsoft, Wednesday, August 26th, 2026
Microsoft Threat Intelligence documents attacks on exposed AI workloads including LiteLLM gateway exploitation and cryptomining.
Microsoft Threat Intelligence examines attacks against exposed AI workloads, where the infrastructure supporting AI has itself become the target rather than the data it processes.
The analysis covers exploitation of LiteLLM gateways, which sit as control points in front of model endpoints and are frequently deployed without the hardening applied to other internet-facing services.
Attackers use that access for credential harvesting, harvesting API keys that unlock paid model capacity, and establish persistence. Observed follow-on activity includes cryptomining on the GPU capacity provisioned for inference. The post covers securing these gateways and control points.