Agent Role Governance

Regenemm treats AI assistants as bounded workflow participants, not autonomous clinicians.

Agent role governance defines what an assistant may do, what it must not do, what data it may access, which tools it may use, when it must escalate, and where human review is required.

The operating principle is simple:

AI assists.
Clinicians decide.
Patients control sharing through Regenemm Link.
Audit persists through the Hub.

Why roles matter

Healthcare workflows are not generic task queues. A documentation assistant, triage assistant, billing assistant, medicolegal assistant, interoperability assistant, and patient-education assistant carry different risks.

Each role must be defined before an agent can safely operate.

Role definitions help prevent:

• silent changes to clinical truth;
• tool use outside the declared workflow;
• over-broad access to patient context;
• unreviewed patient-facing release;
• unclear responsibility between AI, clinicians, care teams, administrators, and patients;
• audit gaps after an agent action.

Role definition model

Each agent role should declare:

• role name;
• workflow purpose;
• parent Hub or spoke surface;
• permitted data classes;
• forbidden data classes;
• permitted tools;
• forbidden actions;
• human owner or reviewer;
• escalation triggers;
• release gates;
• audit events;
• runtime and network boundaries.

This makes agent behaviour inspectable at the workflow level.

Example role classes

Regenemm may use role classes such as:

• Documentation assistant: prepares draft clinical artifacts from approved consultation evidence.
• Patient-summary assistant: prepares plain-language summaries for review and release.
• Education assistant: drafts condition-specific education that remains aligned to the consultation.
• Triage assistant: identifies missing information, priority signals, and escalation preparation.
• Billing assistant: prepares billing-readiness context without becoming clinical truth.
• Medicolegal assistant: prepares chronology and evidence packages with stricter provenance controls.
• Interoperability assistant: supports mapping, import, export, and system-boundary checks.
• Observability assistant: reviews operational signals without exposing patient-identifiable content.

These are role boundaries, not clinical licences.

Required human review

Human review is required before material clinical reliance, patient-facing release, external disclosure, medicolegal use, billing reliance, or durable change to clinical state.

The review requirement should be visible in the workflow, not implied by policy text alone.

Where review cannot be completed, the action should fail closed.

Hub-governed agent activity

Regenemm Voice is the Hub and clinical control plane.

Agent actions should return to Hub-governed pathways for:

• clinical state;
• evidence links;
• source provenance;
• audit events;
• release gates;
• consent state;
• data permanence.

Spokes contextualise work. The Hub governs and persists.

Agent actions that are not permitted

Agentic assistants must not:

• make final clinical decisions;
• impersonate a clinician, patient, carer, administrator, or external system;
• bypass Hub governance;
• silently mutate clinical truth;
• access patient data outside declared workflow need;
• release patient-facing material without an authorised release gate;
• contact patients without an approved workflow;
• use identifiable patient data for foundation model training by default.

Public disclosure posture

This public resource explains Regenemm's role-governance model at a safe level.

Detailed internal role manifests, tool permissions, infrastructure controls, test evidence, security implementation detail, and clinical safety sign-off remain internal governance artifacts.

Summary

Agent role governance makes AI assistance safer by giving every assistant a declared purpose, boundary, reviewer, evidence path, and audit trail.

Regenemm's position remains:

Agents assist.
Humans review.
The Hub governs.
Audit persists.