An inference service is a system that hosts a machine learning model and returns predictions, classifications, scores, or structured output when it receives data. Unlike model training, inference is the operational phase: the model is already built, and the service is used to answer requests quickly and repeatedly.
In cyber security, inference services matter because they can automate decisions. Defenders use them to score phishing messages, flag anomalies, or enrich alerts. Attackers can also abuse them by sending telemetry or content to a model endpoint and turning the result into follow-on actions, task labels, or routing decisions. Analysts should watch for suspicious outbound API calls, serialized host data, and queueing patterns that convert raw observations into machine-readable instructions. Those signals can reveal a malicious control workflow hiding behind ordinary AI integration.