Operation is the running life of the model: serving predictions, watching behaviour and data, and deciding when a retrain or a withdrawal is needed.
What the base framework establishes
The clause follows deployment. The model is available as a stand-alone application or as a function inside another system, and it can be used in batch mode or in a live flow. Operation is where the assumptions made in development meet production data, which may drift from the data the model learned. The framework's cross-cutting concerns apply here as they do at every stage, and the pipeline does not end at go-live. The clause text is not in the public sample beyond the opening, and this page states the stage's role in the framework rather than quoting it.
Why it matters to an AI management system
A model is a living component, not a delivered file. Operation is where the evidence of continued fitness is produced: monitoring, drift checks, incident records and the decision to retrain. An AI management system that has no operation evidence has no answer when the auditor asks how the model is kept fit for purpose.
The public draft summary names no change to 8.7 Operation. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Monitoring record | The signals watched, their thresholds and the response when one fires. |
| Drift report | Input and output drift against the development baseline. |
| Retrain decision | The trigger, the decision and the evidence for it. |
One question to prepare for. What would tell you the model has stopped being fit for its purpose?
Common gaps
- Only uptime is monitored, not the data or the predictions.
- Drift is detected but no one owns the decision to act.
- A retrain happens silently and the model record is not updated.
Prepare before publication
Define the monitoring signals and thresholds, name the owner of each, and record every retrain decision against the model record.
Dates that matter
No separate date attaches to 8.7; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
8.5 Verification and validation · 8.6 Model deployment · 8.8 Example machine learning process based on the pipeline
Source
Source for 8.7: ISO/IEC 23053:2022, clause 8.7. The public sample stops before 8.7, so this page relies on the public contents and the public summary and does not quote the clause. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.