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.
Policy changes are a poor excuse for leaving a product decision unmade. Some questions need a legal interpretation. Others need the team to decide what the system is allowed to do. Confusing the two can keep both waiting.
Separate the decisions that are waiting
Consider an agent that can draft a recommendation but cannot act on it. The team is considering adding execution permissions. It may need legal advice about the new use. It also needs a product decision about scope, an engineering design for enforcing that scope and tests for failure.
Put those dependencies in separate rows. “Waiting for guidance” should identify a specific unresolved question and the decision it blocks. It should not become a blanket status for work that could already be done safely.
This does not mean ignoring the law or building around it. It means preserving the distinction between a legal unknown and an unfinished design. Both deserve attention; neither becomes clearer when they share one vague label.
| Decision | Evidence to request |
|---|---|
| Legal interpretation | Specific question and qualified reviewer |
| Product scope | Permitted and excluded behaviours |
| Engineering validation | Tests of the proposed boundary |
The organizations building AI accountability that actually holds up are not doing it because a regulator told them to. They are doing it because they have understood that accountability is a product quality problem. It is about what your model does in the hands of a real user, in a real situation you did not fully anticipate. No framework written in 2024 is going to resolve that for you.
The regulation is coming. Some of it is already here. But the work of building AI you can actually stand behind is different from the work of demonstrating compliance, and the distance between those two things is where most of the risk lives.
What to do with this
Review what can progress without prejudging the unresolved legal issue. Documentation of the current workflow, reversible test-environment experiments and a clear inventory may still be useful. Keep any dependency on legal advice explicit rather than guessing the answer.
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.
