- Private AI
- AI workloads run inside an approved environment and data boundary rather than being sent indiscriminately to public services.
- Sovereign AI
- AI infrastructure, data handling and governance designed to remain under the required Australian operational and legal controls.
- On-premise AI
- Models and retrieval systems that run on infrastructure physically controlled by the operator.
- Air-gapped
- A system isolated from public networks so protected records and workloads cannot leave through an internet path.
- Inference
- The live step where a model produces an output from an input and the context supplied to it.
- Inference topology
- The layout of model runtimes, data stores, policy controls, audit logs and approved routes that determines where each AI workload runs.
- Retrieval-augmented generation
- A pattern where the system retrieves relevant approved material before composing an answer, keeping evidence close to the output.
- Fine-tuning
- Adapting a model to repeated terminology or task patterns while keeping retrieval, permissions and evaluation as separate controls.
- Provenance
- The recorded path from an AI output back to its inputs — which documents, which model, which approvals. Shows where an answer came from; a person still decides whether it's right.
- Operational knowledge
- The records a business runs on — project files, contracts, procedures, correspondence — made searchable by AI without changing who may see what, which revision counts, or how each document is classified.