Governance as state transitions, not documents
Most AI governance material lists principles. Principles are necessary but unenforceable: they do not tell an engineer what cannot ship today. This framework instead attaches controls to the moments where an AI system changes state, because those are the only moments where a control can actually stop something.
The result is a small number of gates that a delivery team can satisfy without a separate governance programme, and that an auditor can test after the fact from artefacts the team produced anyway.
The six control points
- Intake — the decision the system supports, its owner, and the harm cases it must refuse
- Data access — entitlement-aware retrieval, retention limits and lawful basis for each source
- Evaluation — a curated set labelled by practitioners, re-run on every model or prompt change
- Release — model card, known failure modes, rollback path and human review route
- Operation — logging of inputs, outputs and overrides, with drift and cost monitoring
- Retirement — decommissioning, data disposal and the record of what the system decided
Artefacts that carry the evidence
Each control produces one artefact, and each artefact has a single owner. The intake record names accountability; the evaluation set converts opinion into a repeatable check; the model card documents behaviour and limits; the run book defines what an operator does when confidence falls or a source goes stale.
Where an organisation already runs change management or model risk processes, these artefacts are designed to slot into them rather than replace them.
Mapping to regulation
The framework is regulation-neutral by design. Sector rules typically raise the bar on two controls — evaluation evidence and operational logging — rather than adding new stages, so the model is extended by strengthening those gates rather than by restructuring the lifecycle.
Licence and citation
Free to read, quote and cite with attribution to SJAIN Tech. Redistribution of the PDF in modified form is not permitted.
Cite as: SJAIN Tech (Sjain Ventures Ltd), “A practical AI governance framework: control points from intake to retirement”, 4 August 2026. https://tech.sjain.io/research/ai-governance-framework-reference-model