Treat the agent as a principal, not a new legal category
An AI agent that queries customer accounts can be handled with the access concepts the Rule already uses: it is a system or principal that touches customer information, so the same expectations apply to it as to other users and systems.
In practice, the interpretation translates into four things:
- Distinct agent identities
- Scoped permissions
- Tool-call authorization
- Audit trails
What the access expectations mean when the actor is an agent
Access to customer information should be limited to authorized users, restricted by need, and logged and monitored. When the actor is an agent, each of those themes maps to a specific runtime control.
Mapping Rule themes to runtime controls
| Rule theme | What it means for an agent |
|---|---|
| Authorized access | Each agent has its own identity, tied to an accountable owner. |
| Need-based restriction | Permissions are scoped to the task and the customer information it requires. |
| Tool-call authorization | Each call is evaluated against policy before it executes. |
| Logging and monitoring | Records link the agent, the request, the action, and the decision. |
Where shared credentials and dynamic tool calls undermine evidence
A shared service credential makes it hard to show which agent acted, and tool calls that are not authorized against policy before they run leave gaps in the record. Either weakness makes it harder to produce audit-trail evidence of access to customer information.
Assess where your current agent deployments fall short
For each agent deployment, ask the following questions:
- Does the agent have its own identity rather than a shared service credential?
- Are its permissions limited to the customer information its task requires?
- Is each tool call or API call authorized against policy before it runs?
- Can each logged action be tied to the agent identity and to the requesting user or process?
- Does a named owner review the agent's access rights on a defined schedule?
- Could you produce audit-trail evidence of its access to customer information if a regulator or auditor asked?
Where runtime governance fits
Runtime governance is where the expectations above become evidence. Each agent call is evaluated against policy before it executes, and the resulting records link the agent, the request, the action, and the decision, so access to customer information can be reviewed after the fact.