The system is layered: applications are built from models, models solve tasks, and both depend on tools and data. One system can hold several models.
What the base framework establishes
The clause describes the elements of an ML system and the roles around them, noting that the functions can sit with different entities, including different vendors. Applications are built from models; models solve tasks; both depend on tools, techniques and data. A single system can combine several models. Components can be described by input, output and intent, and can be tested on their own. A pre-trained model is one already trained when obtained, and transfer learning adapts such a model to a related task. The word application covers both the intended use and the software that implements it.
Why it matters to an AI management system
This is where vendor boundaries become visible. If a foundation model is consumed as a service, the model component belongs to the provider while the application and the task belong to the buyer. The description must say which is which, or accountability blurs exactly where a regulator will look.
The public draft summary names no change to 6.1 Overview of the machine learning system. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Component register | Each model, application and tool, with its provider and owner. |
| Vendor boundary note | Which components are consumed as a service and what the provider supplies. |
| Component test record | The input, output and test for each component that can be tested alone. |
One question to prepare for. Which components of your system are you actually accountable for?
Common gaps
- The description treats a third-party model as if the organisation trained it.
- Several models are in one application and the register lists only one.
- Pre-trained and fine-tuned components are not distinguished.
Prepare before publication
Build a component register with a provider column and a trained-here column. Use it as the spine of the AI system description.
Dates that matter
No separate date attaches to 6.1; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
1 Scope · 5 Overview · 6.3 Model · 8.6 Model deployment
Source
Source for 6.1: ISO/IEC 23053:2022, clause 6.1. The statement is read from the public base sample at 6.1 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.