Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates and primary reference.
Self-assessment is still assessment. The useful question is what evidence supports the conclusion and what would cause it to be revisited. A label selected in a form should not end the conversation.
Make classification reviewable
A classification record should explain the system’s purpose, relevant context and reasoning. Keep the source of the criteria visible and identify who reviewed the interpretation. The conclusion needs to be understandable to someone who did not attend the original meeting.
Consider an illustrative assistant whose role changes from summarising information to recommending an action affecting a person. The old assessment may no longer describe the use. A change in purpose deserves review even if the interface still looks familiar.
Record assumptions explicitly. If the classification depends on the system not making a particular decision, connect that assumption to a product boundary. Then test whether the boundary is respected in the actual workflow.
Do not treat this review process as permission to choose the least burdensome category. It is a way to make the reasoning inspectable. Where the legal interpretation is uncertain, seek qualified review and retain the uncertainty in the record.
| Decision | Evidence to request |
|---|---|
| Reasoning | Purpose and applicable criteria |
| Boundary | Assumptions enforced in the product |
| Reassessment | Changes that invalidate the conclusion |
What to do with this
The best classification record is one that can be challenged productively. A reviewer should be able to point to the assumption they disagree with, rather than debate a label without its reasoning.
Record the decision, the responsible owner and the evidence still missing. The decision-record example shows a lightweight way to do that. Where a claim concerns system behaviour, connect it to evaluation evidence. These are proposed product practices, not a finding that a system complies with the law.
Primary timeline reference: European Commission, AI Act. The operational recommendations above are my analysis.
