A model is a mathematical construct plus software that turns input into an inference or prediction. It is trained to represent patterns in data, and it carries those patterns, including their flaws.
What the base framework establishes
The clause defines an ML model as a data structure and the software that processes it, both fixed by the chosen algorithm. Inputs and outputs are set for the model, which is then populated or trained so that it represents the statistical properties of the training data. Its output is an approximation, produced by optimisation rather than by exact solution. The clause makes a point that matters to governance: the model is a result of its training data, so incomplete data or data that carries societal bias will show up in performance. Care with the datasets is therefore part of model quality, not a separate concern.
Why it matters to an AI management system
This is the sentence to quote when someone treats a model as a neutral calculator. The framework itself says the model reflects its data. That gives the governance lead a standard-backed reason to require a data statement and an evaluation on data the model has not seen.
The public draft summary names no change to 6.3 Model. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Model card | The algorithm, inputs, outputs, training data and intended use. |
| Training record | The training data version, the date and the configuration. |
| Performance note | What the model approximates and where the approximation is weakest. |
One question to prepare for. What does this model reflect about the data it was trained on?
Common gaps
- The model card lists the algorithm but not the training data version.
- Performance is claimed without a statement of the population the data represents.
- The intended use is absent, so the model is reused outside its design.
Prepare before publication
Add intended use, training data version and the known limitations to the model card. Have the model owner sign it.
Dates that matter
No separate date attaches to 6.3; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
3.1 Model development and use · 6.4 Data · 8.4 Modelling · 8.5 Verification and validation
Source
Source for 6.3: ISO/IEC 23053:2022, clause 6.3. The statement is read from the public base sample at 6.3 and is written in our words. 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.