Inventory the generative systems, write the description in the framework vocabulary, record the architecture, and add a synthetic data rule. The certifiable work stays in ISO/IEC 42001.
Start with the inventory and the description, because both are useful whatever the final wording says. For each generative system, record the task, the model family, the data, the tools and the architecture. Then write a synthetic data rule with a share limit and a model collapse check, because the draft treats generated training data as a named risk. Refresh the 42001 impact assessment and statement of applicability, since those are the auditable obligations. Ask the certification body or auditor in writing which vocabulary it will sample at the next audit, and record the reply. Re-date the plan when ISO publishes, and watch the parallel 22989 amendment and the AWI Amendment 2 project.
What is the first artefact?
The inventory of ML-based systems with the generative ones marked. Without it, every later step is guesswork.
What is the most common mistake?
Treating the amendment as a certification deadline and waiting for the published text. The useful work starts now and does not depend on the final wording.
Where this comes from
The clause record states each unit in our own words and cites the source. The amendment page separates the public draft summary from the published base edition. The amendment record is ISO standard 88149.
Not legal advice; does not certify. The amendment wording is not held and is not stated.