What a Health System Should Do When a Clinical AI Vendor Is Acquired or Shuts Down
Treat the event as a governance and continuity problem, not only a procurement issue. Keep a current inventory and contract protections in place beforehand. When a signal appears, assess exposure, contain access and data flows, then migrate or decommission with clinical sign-off, verified data deletion, and preserved audit records.
Why this is a continuity event
A clinical AI vendor acquisition or shutdown is a governance and continuity problem, not only a procurement issue. The effects reach clinical safety, PHI, access control, and contracts, so the response belongs with AI governance leaders as well as procurement.
Six phases of a vendor acquisition or shutdown response
The response moves through six phases. The sections that follow cover each one in order.
| Phase | Focus |
|---|---|
| 1. Preparation | Inventory tools, dependencies, owners, and fallback procedures. |
| 2. Trigger detection | Recognize the first signal and route it to governance. |
| 3. Assessment | Establish what changes for clinical safety, PHI, and contracts. |
| 4. Containment | Limit access, credentials, data flows, and unapproved changes. |
| 5. Transition or decommissioning | Migrate or retire the tool and confirm data disposition. |
| 6. Post-event review | Update the playbook, inventory, and contract templates. |
Preparation: inventory and fallback before any signal
Keep a current inventory and contract protections in place before any signal appears. Inventory tools, dependencies, owners, and fallback procedures.
Contract terms to seek
- Advance written notice of any change of control, with a defined notice period.
- Notice of changes to model versions, hosting location, subprocessors, or data handling.
- Return of PHI and model-related data on termination, with written certification of deletion.
- Transition assistance and continued support for a defined period after notice.
- Access to model documentation and validation artifacts, with escrow where feasible.
- Termination or review rights that are triggered by a change of control.
Trigger detection and assessment
Trigger detection
Recognize the first signal and route it to governance.
Assessment
When a signal appears, assess exposure. Establish what changes for clinical safety, PHI, and contracts.
Containment: limit access without breaking care
Contain access and data flows while the tool is under review. Limit the following without breaking care:
- Access
- Credentials
- Data flows
- Unapproved changes
Transition, decommissioning, and post-event review
Transition or decommissioning
Migrate or retire the tool and confirm data disposition.
Post-event review
Update the playbook, inventory, and contract templates.
Keep control of access while a clinical AI tool is under review
Runtime policy enforcement, agent identity, least-privilege permissions, and audit logging can help limit access and preserve records during a vendor transition. Talk through how these controls fit your governance process.
Talk to an Expert