AI operating boundaries
The B.O.R.D.E.R. methodology helps you decide what AI and AI digital workers may do on their own and where a human decision is needed. It also helps you decide who is accountable for it and how to document and review the agreed boundaries.
- Intent
- Input control (applies)
- Authorization (applies)approves the operation
- AI
- Result validation
- Human
- Evidence (applies)
Highlighted: the points of the flow that this description refers to. Orange marker: a human decision at this point.
What problem the methodology addresses
AI prepares documents, uses tools and acts as an AI digital worker. The rules for such work are often written down only in a general policy, or not at all. In practice, three questions remain unanswered.
What can AI do without approval?
It is unclear which actions an AI digital worker performs on its own and which require a decision by an authorized person.
Who is accountable for the result?
There is no named owner of the process, of its decisions and of the AI digital worker acting on its behalf.
What if a boundary is crossed?
Nobody knows who receives the case, who decides it and whether work should stop.
What the procedure looks like
The six B.O.R.D.E.R. steps lead from describing the boundaries in a process to reviewing them regularly. The letters of the acronym come from the English names of the steps.
Step 1 · B (Boundary)
Decision boundaries
For each process, describe which actions and decisions AI may perform and which remain with a human. The boundary is set for a specific process, not for AI in general.
Step 2 · O (Ownership)
Owner and accountability
Name the process owner and the people authorized to approve. Assign each AI digital worker a person who is accountable for the scope of its actions.
Step 3 · R (Risk)
Impact of errors
Assess what could happen if AI makes a mistake in a given step: for the customer, the data, the finances and legal obligations. This assessment determines where approval is needed. The organization carries out the legal classification of the use case separately.
Step 4 · D (Documentation)
Documenting the rules
Record the agreed boundaries, approval thresholds and roles so that they can be reviewed and updated. Decide which events are to be recorded and how long they are kept.
Step 5 · E (Escalation)
Escalation
Decide what happens when an action goes beyond the boundary: who receives the case, who makes the decision and whether work waits for that decision.
Step 6 · R (Review)
Review
Review the boundaries regularly based on recorded events and reports. Check whether the thresholds are right and whether AI digital workers have broader permissions than the task requires.
Boundaries in an AI digital worker’s work
- Step 1: Boundaries for the AI digital workerThe organization sets the AI digital worker’s maximum H.A.D.A. zone, its approval threshold and its operation and cost limits.
- Step 2: Acting within the boundariesThe AI digital worker may perform operations that fall within the agreed boundaries, within the scope of the tools granted to it.
- Step 3: Decision above the thresholdOperations above the agreed threshold require a human decision before execution.
- Step 4: Review and responseRecorded events are the basis for reviewing the boundaries; if needed, the AI digital worker is suspended or revoked.
Which steps of the method AiTrust supports
A methodology describes how an organization sets the rules for working with AI; AiTrust provides technical support for selected steps, to an extent that depends on the deployment configuration.
Agent passport
- By default in AiTrust
- Without a grant and an active agent passport, the AI digital worker will not use tools, and the passport can be suspended or revoked.
- Your organization decides
- The organization sets the AI digital worker’s maximum H.A.D.A. zone, its approval threshold and its operation and cost limits.
Related steps
Decision boundaries (B), Escalation (E)
Human decision before execution
- By default in AiTrust
- In the standard profile, an operation in zone C waits for approval by an authorized person other than the requester.
- Your organization decides
- The organization decides which classes of operations require a human decision before execution.
Related steps
Decision boundaries (B), Owner and accountability (O), Escalation (E)
Document permissions
- By default in AiTrust
- The baseline policy of the Data area for your sector has an initial configuration of document permissions and of the data check before sending to the model.
- Your organization decides
- Which documents are available when working with AI depends on the assigned permissions and on how the sources are configured in the given deployment.
Related steps
Decision boundaries (B)
Evidence trail
- By default in AiTrust
- Operations in the controlled path leave an evidence trail that can be used to review the agreed boundaries.
- Your organization decides
- The organization sets the evidence retention period.
Related steps
Documenting the rules (D), Review (R)
How this step looks in the AiTrust application
In the AiTrust application, an AI digital worker acts as an agent on the basis of an agent passport. Its operating boundaries are visible in the passport registry. The registry shows the status, the maximum H.A.D.A. zone and the approval threshold of each passport.
What the available example shows
The agent passport is the place in AiTrust where the operating boundaries of a specific AI digital worker are recorded.
What this example shows
- The passport status “Active” and the “Suspend passport” and “Revoke passport” buttons.
- The actor type ai_agent and the passport’s issue and expiry dates.
- The maximum H.A.D.A. zone B and approval required above zone B.
- A limit of 100 operations per hour and a daily cost limit.
- The tool grants and delegations tabs (with no entries in this example).
What this example does not show
The screenshot shows the interface and the configuration of one AI digital worker on synthetic data. It does not prove that the limits work the same way in every deployment, and it does not show other AI digital workers or cases.
Related answers on the platform and compliance
Answers on the platform page and on the AI Governance and compliance page
- Actions of AI digital workers
Which actions an AI digital worker may perform on its own and which require a decision by an authorized person.
- Human decision in the process
How consent to perform an action differs from approval of a document’s content.
- Record of events and decisions
What the scope of the record and its retention period depend on.
- AI Act
Information on the AI Act in the security and compliance area.
Other methodologies
- Human oversight of AIH.A.D.A.
- Readiness assessment methodologyI.A.G.R.A.
- Protection and resilienceS.H.I.E.L.D.
Let’s talk about AI operating boundaries in your processes
In the demo, we will show the agent passport and action approval in a demonstration configuration. We will also discuss which boundaries are worth setting for AI digital workers in your organization.