• +1 (602) 922-5990
  • This email address is being protected from spambots. You need JavaScript enabled to view it.
  • Mon - Fri 8:00a - 5:00p AZ
Industry Insights banner: Governing Vendors and Clinical AI, a resilience agenda for healthcare, over a mountain valley at dusk.

Governing Vendors and Clinical AI: A Resilience Agenda for Healthcare

A health system's continuity plan can name generators, evacuation routes, and clinical documentation and still leave a quieter gap: the electronic health record, the imaging archive, the staffing platform, or a clinical decision tool that is owned, hosted, or updated by someone else. When one of those services fails, the event is not only an information-technology outage. It can become a care-continuity, privacy, and AI-governance event at the same time.

Healthcare already has official requirements that touch this gap, even if they were not written as an "AI and vendor resilience" program. The Centers for Medicare & Medicaid Services emergency preparedness condition of participation for hospitals, at 42 CFR 482.15, requires an all-hazards program, a plan reviewed at least every two years, and exercises at least twice a year. The HIPAA Security Rule requires covered entities and their business associates to safeguard the confidentiality, integrity, and availability of electronic protected health information, including through a contingency plan. Neither rule is a complete method for governing clinical AI or a complex vendor estate. Both are reasons not to treat those issues as optional.

This article is about that combined agenda: how hospitals and health systems govern contracted services and AI tools so that care can continue when a vendor or a model fails. It is a reading of the official requirements as they now stand, and of the questions leaders should ask of the third parties and models that sit in the path of care.

Why this matters

Healthcare operations depend on arranged services: electronic health records, revenue-cycle platforms, laboratory and pharmacy systems, imaging, telehealth, medical devices, staffing, cloud hosts, and AI tools used for documentation, imaging support, scheduling, or clinical decision support. Some of those partners are business associates under HIPAA. All of them can interrupt care if they cannot perform.

The CMS emergency preparedness rule makes that dependency visible in plain regulatory language. For hospitals, 42 CFR 482.15(c) requires the communication plan to include contact information for "entities providing services under arrangement." The training program in 42 CFR 482.15(d)(1) must include initial training for "individuals providing services under arrangement," consistent with their expected role, as well as for staff and volunteers. The plan itself must address continuity of operations, including delegations of authority and succession plans, and policies must address a system of medical documentation that preserves patient information, protects confidentiality, and maintains the availability of records. Those are continuity and third-party requirements, written into a condition of participation.

The HIPAA Security Rule, at 45 CFR Part 160 and Subparts A and C of Part 164, requires administrative, physical, and technical safeguards to ensure the confidentiality, integrity, and availability of electronic protected health information. Under 45 CFR 164.308(a)(7), covered entities must establish, and implement as needed, a contingency plan for emergencies that damage systems containing ePHI. Required specifications include a data backup plan, a disaster recovery plan, and an emergency-mode operation plan. Addressable specifications include periodic testing and revision, and an applications-and-data criticality analysis. Business associate arrangements are part of the same administrative-safeguard structure. Availability is a Security Rule objective, not a vendor courtesy.

HHS published a proposed update, HIPAA Security Rule to Strengthen the Cybersecurity of Electronic Protected Health Information, on January 6, 2025. As of HHS's Security Rule page (content last reviewed March 19, 2026), that proposal is still presented as a proposed rule. The current Security Rule remains the requirement. Organizations should track the proposal. They should not defer backup, recovery, emergency-mode operations, or business-associate oversight while they wait for a final rule.

What the emergency preparedness rule already asks about partners

CMS published the Emergency Preparedness Requirements for Medicare and Medicaid Participating Providers and Suppliers final rule in the Federal Register on September 16, 2016 (81 FR 63860), with later revisions in a September 30, 2019 burden-reduction rule (84 FR 51817). The requirements apply to multiple provider and supplier types, each with its own regulatory section. The hospital text in 42 CFR 482.15 is the version discussed here; other provider types should read their own section.

The hospital must maintain a comprehensive, all-hazards emergency preparedness program. The emergency plan, policies and procedures, communication plan, and training and testing program must each be reviewed and updated at least every two years. Testing must include exercises at least twice per year: an annual full-scale community-based exercise or, if that is not accessible, an annual facility-based functional exercise, plus an additional annual exercise that may be a second full-scale or functional exercise, a mock disaster drill, or a facilitated tabletop. Activation of the plan for an actual emergency exempts the hospital from its next required full-scale or functional exercise.

Those mechanics define how often the organization must look again at its plan and prove that it works. A program that never exercises the loss of an electronic health record, a cloud-hosted imaging archive, a staffing platform, or a clinical decision tool is not testing the disruptions most likely to affect modern care. The rule does not name those scenarios. It does require a facility-based and community-based risk assessment, strategies for the events identified, and analysis of every exercise and emergency, with plan revisions as needed. If arranged services and AI tools are material to continuity of operations or to the availability of records, they belong in that assessment and in the exercise program.

The rule also allows a healthcare system with multiple separately certified facilities to use a unified and integrated emergency preparedness program, provided each facility participates, the program accounts for each facility's circumstances, and each facility can actually use it. That option is useful for vendor and AI oversight only if the unified program can see shared platforms and shared models, not only shared policies.

Clinical AI and the vendor behind the model

Health systems are adopting AI for documentation, imaging support, revenue-cycle work, and, in some settings, clinical decision support. Some of those tools are medical devices. Some are not. Some are developed and updated by the organization. Many are supplied, hosted, and retrained by a vendor. In every case, a model that supports care or the availability of health information needs a defined intended use, a view of data quality and limits, human oversight, monitoring after deployment, and a plan for what happens if the model is wrong, unavailable, or withdrawn.

The National Institute of Standards and Technology published the Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, on January 26, 2023. It is voluntary and organized around Govern, Map, Measure, and Manage. It asks organizations that design, develop, deploy, or use AI to manage risk across the lifecycle and to pursue trustworthy AI, including systems that are valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair, with harmful bias managed. Those characteristics map directly onto care quality and onto the Security Rule's confidentiality, integrity, and availability objectives.

ISO/IEC 42001:2023, published in December 2023, is the first international management-system standard for AI, with requirements for establishing and continually improving an AI management system in organizations that provide or use AI systems. Certification is voluntary and is not performed by ISO. For a health system, it offers a single set of policies, roles, and lifecycle controls covering internally developed tools and vendor-supplied models.

Organizations that place or deploy AI in the European Union also need to account for Regulation (EU) 2024/1689, the EU Artificial Intelligence Act, which entered into force on 1 August 2024 and applies in phases. Most remaining obligations have applied since 2 August 2026, with some high-risk AI rules applying later. Whether a particular clinical or administrative tool is in scope is a legal and factual question, especially where the tool may also be a medical device. The operational point does not depend on that classification: if a model can affect care, privacy, or the availability of records, the organization should be able to explain its purpose, its limits, who may change it, and how it will be taken out of service without interrupting care.

The US Food and Drug Administration has also issued draft, non-binding guidance on AI, including the January 2025 draft, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, which is marked not for implementation, and draft recommendations on AI-enabled device software functions. Those documents are directed primarily to sponsors and device developers. They are useful reading for a health system that depends on such products, but they are not a hospital governance program.

Where vendors, AI, and continuity meet

The meeting point is a service the organization cannot pause without affecting patients: access to records, medication and laboratory information, scheduling and staffing, imaging, or a decision tool used in a clinical workflow. The vendor's incident, or the model's withdrawal, becomes the hospital's continuity problem. HIPAA business-associate terms and CMS communication-plan contacts are necessary. They are not a substitute for knowing whether the provider can recover within a tolerable time, whether a second path exists, and whether the model can be disabled without leaving clinicians without a workable process.

That is why third-party oversight and AI governance should use the same map as emergency preparedness and the Security Rule contingency plan. The applications-and-data criticality analysis, the business impact analysis, and the vendor inventory should be readable as one picture. Exercises should include the loss or degradation of a contracted platform or a clinical AI tool, not only facilities and weather. Contracts should address availability, incident notification, data return, audit and information rights, and transition, including for model updates and model withdrawal. Monitoring should include model performance and data drift, not only uptime.

Concentration belongs in that picture. Many hospitals depend on a small number of electronic health record, cloud, and revenue-cycle providers. A failure at a widely used platform is both an individual hospital problem and a community problem, which is why 42 CFR 482.15 already requires a process for cooperation with local, tribal, regional, state, and federal emergency preparedness officials. See also Beyond Due Diligence: Proving Third-Party Resilience, The Hidden Risk of Concentration, and When a Critical Provider Cannot Recover.

What organizations should do

Update the emergency preparedness risk assessment and the Security Rule criticality analysis so that arranged services and AI tools material to care or to ePHI are named, owned, and prioritized. Do not keep a vendor list, an AI inventory, and a continuity plan that cannot be read together.

Use the existing CMS review and exercise cycle. At least every two years, confirm that contact information, training, and procedures for entities under arrangement are current. In the twice-yearly exercise program, include at least some scenarios that test the loss of a contracted platform or a clinical AI tool, and document findings with owners and dates.

Align business-associate and other vendor contracts with the backup, disaster recovery, and emergency-mode expectations of 45 CFR 164.308(a)(7), and with the need to train and communicate with people who provide services under arrangement. Require incident notification, evidence of recovery testing, and a workable transition path.

Govern clinical and operational AI with a documented intended use, human oversight, monitoring, and a discontinuation plan. Use the NIST AI RMF and, where useful, ISO/IEC 42001 to structure that work. Do not claim conformity you have not established, and do not treat draft FDA guidance as a final requirement.

Give senior leaders a combined view of care-critical vendors, material AI uses, exercise results, and open issues. See also From Plans to Proven Readiness and Operational Resilience: Where to Start.

How OpResONE can help

OpResONE helps healthcare organizations strengthen operational resilience, third-party oversight, and AI governance as a single management problem, including mapping care-critical dependencies and designing exercises and reports that leadership can use. Our work is independent. We have no affiliation with CMS, HHS, FDA, NIST, or ISO, and this article is not an official interpretation of their rules or guidance.

We bring BRILLIANCE to RESILIENCE®.

Related resources

Request a Strategy Call

If you want a focused reading of how your vendor and AI oversight lines up with emergency preparedness and the availability of care and records, talk with OpResONE about where to start.

Request a Strategy Call

Sources