Inference (AI) is the operational phase in which a trained AI model is used to generate outputs, predictions, classifications, or content, from new input data.
Inference (AI), the operational phase in which a trained AI model is used to generate outputs, predictions, classifications, or content, from new input data.
Inference is distinct from training: training is how the model is built, inference is how it is used in production. Most governance controls that affect end users, input validation, output filtering, logging, human oversight, and latency or cost constraints, operate at inference time. The distinction matters legally too, because obligations such as transparency disclosures and record-keeping attach to the system as deployed and used, not just as developed.
Source: ISO/IEC 22989:2022
Inference is distinct from training: training is how the model is built, inference is how it is used in production. Most governance controls that affect end users, input validation, output filtering, logging, human oversight, and latency or cost constraints, operate at inference time. The distinction matters legally too, because obligations such as transparency disclosures and record-keeping attach to the system as deployed and used, not just as developed.
See where you stand on AI governance
Take the free 7-question maturity assessment and get a personalised action plan.
Free assessment, 3 minutes →