Updated 17 October 2026. Responsible AI has operating costs after the launch deck is finished. Someone has to maintain tests, review exceptions and keep the evidence aligned with the product.
Fund the work that continues
A project budget can cover implementation while leaving the operating model unfunded. The feature ships, but no team has capacity to review failures or refresh the evaluation cases. That is a planning problem even if the original release passed its tests.
An illustrative budget review should separate one-off setup from recurring work. Building a logging path is different from investigating the records. Creating test cases is different from maintaining them as the workflow changes. Both sides need an owner.
Avoid inventing a return-on-investment figure to justify every control. Some decisions can be supported by a clear account of the risk, the operating requirement and the alternatives. Where costs or benefits are estimated, label the assumptions and uncertainty.
Consider scope as a budget lever. A narrower initial action set may be easier to evaluate and support. Expanding autonomy without expanding the operating capacity can transfer the cost to support teams or affected users.
| Question | Evidence or action |
|---|---|
| Setup | Build the control and initial evidence |
| Operation | Review, respond and maintain |
| Expansion | Capacity needed for wider scope |
The practical next step
Ask who will perform the work three months after launch and where that capacity comes from. If the answer depends on spare time, the budget has probably left out part of the product.
Keep the resulting decision, its owner and unresolved questions in a short decision record. For claims about agent behaviour, connect the decision to evaluation evidence rather than relying on a successful demonstration alone.
Further primary reading: NIST AI Risk Management Framework. The review map and product recommendations above are my analysis, not a certification or a legal assessment of a particular deployment.
