Healthcare organizations are beginning to delegate real work to AI agents, from scheduling and prior authorizations to research queries. Each handoff carries a decision about how much authority goes with the task. As an agent adapts its plan, that authority must remain within limits set by the organization. This piece explores how healthcare leaders can establish those boundaries and enforce them as the work happens.
Healthcare work often involves gathering information, coordinating across systems and moving a request toward a decision. Delegating parts of that work to AI agents offers a way to reduce administrative effort. It also requires healthcare leaders to decide how much authority an agent needs to complete the assignment and where that authority ends.
Consider an agent assembling a prior authorization request. When it discovers that a supporting lab result is missing, it could search another system, ask a staff member for help or pause the request. Each option has different implications for access to patient information and responsibility for the next step.
Approving that workflow means deciding which records the agent may access, which actions it may take and when a person must approve its next step. Those decisions define the agent’s authority. They warrant the same scrutiny as any decision about who can see or change patient information.
Every agent deployment involves three decisions. First, what information may it access, for example, the supporting records for a particular patient’s authorization request? Second, what actions may it take: assemble documents, draft the request or submit it to an insurer? Third, where is human approval required? Permission to prepare a request does not automatically include permission to submit it.
Clinical governance already handles delegation this way. A scribe or assistant works within a defined scope, and that scope doesn’t widen because the shift got busy. Agents need the same treatment, with one complication.
An agent can revise its plan as it works, identifying missing information and trying another route when the first one fails. That flexibility makes agents useful, but a new step may require access or permission beyond the original assignment.
For the prior authorization agent, retrieving the missing lab result might require access to another department’s records. The missing information may justify a request for additional authority. It does not permit the agent to expand its own access.
For the prior authorization agent, a missing lab result creates a new information need. It does not grant permission to search additional systems. Before the agent’s next request proceeds, a control outside the agent must determine whether the requested access or action falls within the authority the organization has granted.
Confinement and runtime authorization work together to enforce those limits. Confinement defines the agent’s operating environment and the resources it can reach. Independent authorization evaluates each attempted action against organizational policy before allowing it to proceed. Being able to reach a system does not give the agent permission to use everything in it.
Human approval also needs an explicit place in the workflow. If submitting a request requires a staff member’s sign-off, the workflow must prevent submission until that approval is obtained. Any expansion of the agent’s authority requires a separate, authorized decision.
That authority also needs an endpoint. Access granted for an assignment must expire or be withdrawn when the assignment ends, with a record of the actions attempted and the authorization decisions made.
Before agents take on more clinical and administrative work, leadership teams must evaluate these questions:
Who approves what an agent may see and do, and is that approval written down?
Which actions always need a clinician or another named person to sign off?
How is the agent’s access limited to its approved assignment?
When does that authority expire, and who can revoke it before then?
Can we reconstruct which agent accessed a record, on whose behalf and under which authorization?
Healthcare organizations remain accountable for the work they delegate. As agents gain flexibility in how they complete that work, the organization must retain control over the authority they exercise.