GOVERNANCE
An institutional AI workflow needs a clear route for challenge, a temporary decision and a recorded return to use. A fictional committee-pack exercise makes those responsibilities visible.
The review pack is ready. A portfolio summary looks coherent, its charts are readable, and the meeting begins in ten minutes. Then an analyst notices that a liquidity input is older than the portfolio snapshot.
Who can stop the affected conclusion being used? Who decides what the committee receives while the issue is checked? And who authorises a corrected version?
These questions can fall between teams. The analyst may assume the platform owner will intervene. The platform owner may assume the reviewer will spot it. The committee may assume that a result in its pack has already passed the necessary checks. A credible operating process gives the challenge a destination, the temporary decision an owner and the eventual return to use a recorded basis.
The Financial Stability Board's 15 October 2026 final report sets out 12 sound practices that financial institutions are encouraged to consider for responsible AI adoption. Its published list addresses governance and accountability, effective documentation, ongoing performance management and human oversight, among other matters. The exercise below is KADE's editorial proposal for making one handoff concrete. It is not an FSB template, a legal conclusion or a description of a deployed KADE control.
Start with the supported task
In this fictional example, an analyst prepares a portfolio summary to support a committee discussion. The institution should be able to state the question it is meant to answer, the permitted inputs, the intended readers and the existing approval route. That purpose matters when a result travels. A summary prepared for discussion should not quietly become an instruction in another application simply because it is easy to copy.
Write down the handoff from the analyst's workspace to the pack: which source versions were used, which conclusions depend on them and who receives the result. The underlying systems of record remain authoritative. The institution retains the decision.
Give an owner a job they can perform
A short operating sheet can name the accountable owner, the preparing analyst, the reviewer, the technical contact, the receiving workflow and the escalation route. Each person should be able to say what they can actually do when a result is challenged. A name in a document, or an unstaffed mailbox, does not demonstrate that an urgent challenge can reach somebody with authority.
For the stale liquidity input, the analyst flags the affected section and its source. The reviewer checks whether other conclusions depend on it. The pack owner decides what can still be shown to the committee under its existing process. This sequence assigns work without pretending that every institution needs the same hierarchy.
Define a usable pause
The proposed exercise gives staff three reasons to hold an affected output for review: a stale material input, an unexplained change in a material result, or a mismatch between the source permission and the intended recipient. The institution would set its own materiality criteria. There is no universal numerical threshold in this example.
A hold should have a scope. If only one conclusion depends on the stale input, a notice can identify that conclusion as unavailable while separately approved information continues through the normal process. If the dependency cannot be isolated, the whole summary may need to wait. In either case, the committee needs a clear account of what is known, what is being checked and when it will hear more.
The ten-minute version of the exercise is deliberately uncomfortable. The analyst identifies the stale input; the reviewer traces the affected conclusion; the pack owner substitutes a short limitation notice; and the committee follows its own escalation and decision arrangements. No automated investment action is implied.
Make return to use a separate decision
A successful rerun is evidence, but it is not the whole decision. The reviewer also needs to know what changed, whether downstream copies require correction, whether the replacement answers the original question and who authorises it to be used. The challenged version and the correction should remain linked under the institution's adopted records policy. This article does not prescribe a retention period.
The return decision is easier to examine later when it records the source revision, the affected output, the reason for the hold, the review outcome and the person who released the replacement. If any of those facts cannot be recovered, the exercise has found a design gap.
Rehearse before relying on it
Run the scenario with synthetic information. Remove an input, alter a recipient's permission or introduce an unexplained difference between two versions. Observe whether staff recognise the issue, find the owner and reach a temporary decision. Record unanswered calls and unclear handoffs, then revise the operating sheet and try again.
A successful tabletop exercise proves only what was tested. It does, however, expose practical questions before a real committee is waiting for an answer. That is the useful point of the exercise: give people an imperfect result and watch what happens next.
Source and scope. The governance context is the Financial Stability Board's final report and published sound-practice list, 15 October 2026. The owner sheet, pause examples and rehearsal are KADE's proposed operating exercise. Applicability to any particular institution requires its own assessment. KADE's current platform status distinguishes the synthesis core in staging from the specified lens suite; this article makes no claim that a pause mechanism has been implemented or validated in KADE.
Illustrative image generated with AI.