Ongoing governance

From an AI inventory to ongoing governance

Your inventory has identified a workflow that uses a model and interacts with a business system. That finding does not yet establish who may authorise its actions, where that authorisation is enforced or how the team will find their outcomes.

To move towards ongoing governance, choose a specific operation and define what needs to be verified. The intended result is a scope the team can explain and maintain, with responsibilities and evidence that the rules are applied. It does not require every use to be connected to a single system.

In brief

The question and the useful result.

Question
What must be verified to move from an identified AI use to an operation that is governed over time?
Useful result
A framework of conditions and evidence to distinguish discovered, connected and controlled for a defined operation, and to reassess that scope when things change.

Ongoing governance

Distinguish discovered, connected and controlled

These terms describe different findings. Each needs a stated scope and supporting evidence.

SituationWhat the team can establishWhat this does not establish
DiscoveredA use is identified through a report or a specified source.That its actions are observed or governed.
ConnectedA real exchange with a system is established for a defined operation.That every action passes through the intended rules.
Controlled within the verified scopeEnforcement of an authorisation rule for the relevant action has been observed and documented for the cases and versions examined.General control of agents, other routes or future changes.

A workflow may be connected to read a record while its updates remain outside the verified scope. Keep that distinction visible instead of assigning one overall status to the tool.

Ongoing governance

Conditions to check for an operation

A written rule and a successful connection are not enough. Before describing an action as governed, examine the following conditions together.

ConditionQuestion to resolve
ScopeWhich record, fields and operation are covered? Which other actions are excluded?
ResponsibilitiesWho defines the need, approves permissions and maintains the integration? Who receives a pending decision?
AuthorisationWhich rule applies to this specific action? When is a human decision required?
IntegrationWhere is the rule enforced before the action? Which known paths do not pass through this control?
ObservationWhat records connect the request, decision and attempt to an outcome or a missing confirmation?
MaintenanceWhich changes require another review, and who will carry it out?

Ask for evidence of enforcement in permitted and refused cases relevant to the scope being examined. The presence of an approval button or a valid token does not, by itself, establish that the intended limits are enforced.

Ongoing governance

Example: changing a delivery date

In this illustrative scenario, the inventory lists an agent that proposes delivery dates. A connection allows it to read an ERP record. The team now wants to let it change a date.

That write operation needs its own assessment. Define the records concerned, the permitted change and the person who approves it when approval is required. The integrator needs to show where these conditions are enforced and what happens when they are not met.

Observing the outcome is a separate matter: an acknowledgement may establish receipt of the request without confirming the new date. Agree what evidence will establish the change in the ERP and keep the outcome unknown until that evidence is available.

The scenario does not establish that DUBSAR has an available ERP integration. It helps a team decide what an integration needs to demonstrate.

Ongoing governance

Maintain the scope as the workflow changes

Associate the verified scope with the versions of the workflow, rules and systems examined. Name the person who will keep that information up to date.

An additional editable field, a new action path or a change of owner may require the affected conditions to be reviewed. An earlier check does not automatically cover those changes. The team needs to know what remains established and what needs another assessment.

This governance can draw on existing systems and procedures. The case for an additional component depends on the gaps actually found and the feasibility of integrating it.

To revisit the starting point, use the AI use mapping method. To examine the permissions that need separate decisions, read the framework for models, data and actions.

Ongoing governance

Frequently asked questions

Does every discovered use need an additional control?

No. First define the problem to be solved. Clear ownership or an existing procedure may meet the need. The decision depends on the operation and the limits found.

Does “connected” mean that every action is governed?

No. A connection establishes an exchange within a scope. You still need to verify where authorisation is enforced, which paths are covered and what evidence is available.

Must existing workflows be rebuilt?

Not necessarily. Examine where a decision can be enforced and which responses can update mission state. That work determines the connections required; universal compatibility must not be assumed.

Ongoing governance

Current DUBSAR status

DUBSAR is developing a foundation intended to connect durable mission state, authorisations, human decisions and evidence alongside existing tools. The conditions in this guide help assess a scope; they do not establish that the product has already met them.

Selected workflows have been tested; full B2B integration and validation are still underway. Pilots are not currently open.