Implementing the Joint Commission and CHAI responsible use of AI guidance: a hospital crosswalk
The Joint Commission and CHAI guidance on the Responsible Use of AI in Healthcare (RUAIH) names seven elements: AI Policies and Governance Structures; Patient Privacy and Transparency; Data Security and Data Use Protections; Ongoing Quality Monitoring; Voluntary, Blinded Reporting of AI Safety-Related Events; Risk and Bias Assessment; and Education and Training. It is voluntary guidance for organizations, not for products.
This guide turns the seven elements into owners, evidence, and example controls. It works from the primary text of the guidance (Joint Commission and CHAI, 2025). For background on the certification program, read our certification explainer.
What the guidance is, and what it is not
The guidance is a joint effort of the Joint Commission and the Coalition for Health AI (CHAI). It calls itself "an initial, high-level document" meant to promote a shared understanding of responsible deployment and use of AI. It is guidance. It is not an accreditation standard, and it does not say that a hospital must follow it.
It guides organizations, not products. In its words, it is "not intended to direct the development of AI tools or validate the effectiveness of AI tools themselves." It also says that playbooks will build on it and that "a voluntary Joint Commission Responsible Use of AI certification program will be developed based on these playbooks."
Scope is broad. The guidance defines health AI tools as clinical, administrative, and operational solutions. That includes care support services such as clinical documentation and scheduling, and care-relevant operations such as revenue cycle management, coding, and prior authorization. Your inventory should cover all of them.
The Joint Commission says it reviewed several inputs while preparing the guidance: CHAI frameworks, the National Academy of Medicine AI Code of Conduct principles, the NIST AI Risk Management Framework, and the Bipartisan House Task Force Report on Artificial Intelligence. That is why the crosswalk below maps cleanly to NIST.
The seven elements at a glance
| RUAIH element | What it asks, in plain language |
|---|---|
| 1. AI Policies and Governance Structures | Write policies and stand up a governance body that reports to your governing board. |
| 2. Patient Privacy and Transparency | Protect patient data and tell patients and staff how AI is used. |
| 3. Data Security and Data Use Protections | Secure the data and limit its permitted uses through data use agreements and BAAs. |
| 4. Ongoing Quality Monitoring | Monitor AI tools locally, scaled to risk, with named responsible parties. |
| 5. Voluntary, Blinded Reporting of AI Safety-Related Events | Capture AI safety events internally and share de-identified details with an independent organization. |
| 6. Risk and Bias Assessment | Identify, document, and check for risk and bias before and after deployment. |
| 7. Education and Training | Train users in role-specific ways and document how. |
The crosswalk
Use this table as a working checklist. Each row names the element as the guidance does, who should own it, the evidence to keep, the related rule or framework, and an example control. The guidance does not prescribe owners or evidence formats. These columns are our implementation suggestions.
| RUAIH element | Accountable owner | Evidence to keep | Related rule or framework | Example control |
|---|---|---|---|---|
| 1. AI Policies and Governance Structures | Governance committee chair and executive sponsor | Charter, AI inventory, policy set, board update records | NIST AI RMF GOVERN 1.6 (inventory) and GOVERN 2.1 (roles and responsibilities); CHAI policy and organizational structure domains | Registry of AI tools with a named owner and status for each |
| 2. Patient Privacy and Transparency | Privacy officer | Patient notice policy, consent where relevant, staff transparency materials, disclosure log | HIPAA Privacy Rule; for certified EHR modules, the decision support interventions criterion in 45 CFR 170.315(b)(11) | Log of when and how patients were told AI affected their care |
| 3. Data Security and Data Use Protections | CISO, with privacy and procurement | Signed BAAs, data use agreements, access log reviews, security assessments, incident response plan | 45 CFR 164.504(e) and 164.314(a) (business associate contracts); HSCC Third-Party AI Risk Guide, Appendix F | Least-privilege access scopes with logged access for every AI tool |
| 4. Ongoing Quality Monitoring | Tool owner and quality lead | Risk-scaled monitoring plan, results, vendor change channel, AI dashboard, adverse event path | NIST AI RMF MANAGE 2.4 (supersede, disengage, or deactivate systems) | Drift checks, plus denied-action and stop logs |
| 5. Voluntary, Blinded Reporting of AI Safety-Related Events | Patient safety officer | AI event categories, internal incident records linked to the tool, record of any PSO submission | The guidance names the Joint Commission sentinel-event process and federally listed PSOs | Incident form with an AI event type and a link to the tool record |
| 6. Risk and Bias Assessment | Governance committee with data science | Vendor risk and bias answers, model card, pre-deployment and post-deployment bias review | 45 CFR 92.210(b) and (c) for patient care decision support tools | Subgroup performance review on local data, repeated on a schedule |
| 7. Education and Training | Education lead | Role-specific training records, documented plan for how users receive tool documentation | NIST AI RMF GOVERN 2.2; HSCC Third-Party AI Risk Guide, Appendix H | Completion tracking by role, tied to tool go-live |
Notes on each element
Element 1. The guidance says the governance structure does not need to be a standalone team. It should include a designated individual with appropriate technology expertise and a way to keep the governing body updated on uses, outcomes, and potential adverse events. See our guides on the AI governance committee and the AI agent registry.
Element 2. The guidance says patients should be notified "when appropriate" if AI directly affects their care, and that consent should be obtained "where and when relevant." Policies must cover both patients and staff. For certified EHR modules, our HTI-1 guide explains source attribute transparency and what the HTI-5 final rule (published October 28, 2026) changes.
Element 3. The guidance lists baseline security measures: encryption, access controls (including regular audits of access logs), regular security assessments, and an incident response plan. For data use agreements it lists permitted uses, data minimization, prohibition of re-identification, third-party obligations, and audit rights. Where PHI is involved, it says to execute appropriate Business Associate Agreements. Start with your AI vendor agreements. Our vendor intake guide covers the review steps.
Element 4. The guidance asks for policies that name responsible parties for local monitoring. Its list includes regularly validating and testing tools, evaluating data quality, and "developing an AI dashboard." It also lists a process for reporting adverse events or ongoing errors to leadership and vendors. It says monitoring should be risk-based: tools that inform clinical decisions get checked more often, while administrative tools can be checked periodically. It also calls for feedback channels with vendors about model changes. Read our guide on AI model drift.
Element 5. The guidance asks for confidential, blinded reporting to an independent organization. It names the Joint Commission sentinel-event process and confidential reporting to federally listed Patient Safety Organizations as existing structures, and says to capture AI events in internal incident systems. Treat unsafe recommendations, major performance drops after an update, and biased outputs as patient safety events.
Element 6. The guidance asks hospitals to request vendor information on known risks, biases, and limitations, and to check for bias on local data before and after deployment. It names an AI model card, such as CHAI's Applied Model Card, as a way to collect this. Section 1557 adds a legal layer for patient care decision support tools. Read our Section 1557 guide.
Element 7. At a minimum, the guidance says to define and document how users receive tool documentation and role-specific training. It also suggests AI literacy programs for all staff.
Where agentic AI fits
The guidance does not use the word "agent." Its elements still apply to any AI tool a hospital deploys. Agents take actions, so they add four needs: permission scopes, approval gates, action logs, and a stop control.
OWASP lists this risk as LLM06:2025 Excessive Agency. NIST MANAGE 2.4 asks for mechanisms to supersede, disengage, or deactivate systems that perform inconsistently with intended use. Our guides cover each control: the AI kill switch, least privilege for AI agents, human in the loop, and the tamper-evident audit trail.
A 30/60/90 day plan
- Days 1 to 30. Name an owner for each of the seven elements. Start the AI inventory, including embedded and administrative AI. Charter the governance committee.
- Days 31 to 60. Review BAAs and data use agreements for AI vendors. Write monitoring plans for your highest-risk tools. Open a vendor channel for model change notices.
- Days 61 to 90. Make the AI safety event path live in your incident system. Launch role-based training. Deliver the first update to your governing board.
The voluntary certification: what we know and what to confirm
The guidance says a voluntary certification program will be developed based on the playbooks. Our certification guide summarizes the Joint Commission's June 1, 2026 announcement. Program details can change, so confirm them on the Joint Commission site. Ask directly about:
- the standards or elements of performance for each area
- the review method
- fees
- the timeline
- renewal
CHAI describes its governance playbooks as providing "a framework to achieve the voluntary certification Joint Commission is developing." See the CHAI AI governance workgroup page.
How Skovos fits
Skovos is not affiliated with the Joint Commission or CHAI, and using it does not confer certification, accreditation, or compliance. For connected AI agents, Skovos can supply evidence mainly for elements 1, 3, and 4: a registry with owners and status, logged permission decisions, a hash-chained audit trail, and records of any recall. It does not cover all seven elements. Training, patient notice, bias review, and safety event reporting stay with your teams. See what OVERT evidence looks like in our OVERT guide.
Frequently asked questions
Is the Joint Commission and CHAI guidance mandatory?
No. The RUAIH guidance is voluntary guidance, not an accreditation standard. It describes what responsible deployment and use of AI looks like. Hospitals may still face separate legal duties, such as HIPAA and Section 1557.
Does RUAIH certify AI products?
No. The guidance says it is not intended to direct the development of AI tools or validate their effectiveness. It addresses how a healthcare organization deploys and uses AI.
How do the CHAI playbooks relate to RUAIH?
The guidance says playbooks will build on and operationalize it, and that a voluntary Joint Commission certification program will be developed based on those playbooks. CHAI released governance playbooks in eight domains on May 27, 2026.
What evidence should a hospital keep for RUAIH?
Keep records that show each element working: a governance charter and AI inventory, patient notice policies, signed data use agreements and BAAs, monitoring plans and results, AI safety event records, bias review records, and training completion logs. The crosswalk table lists evidence by element.
Does RUAIH apply to administrative AI?
Yes. The guidance defines health AI tools to include clinical, administrative, and operational tools, including care-relevant operations such as revenue cycle management, coding, and prior authorization. It says monitoring should scale with risk.
How does RUAIH apply to AI agents?
The guidance does not mention agents by name. Its elements still apply to any AI tool a hospital deploys. Agents that take actions add the need for permission scopes, approval gates, action logs, and a stop control.
Does using Skovos make a hospital RUAIH compliant or certified?
No. Skovos is not affiliated with the Joint Commission or CHAI, and no software confers certification. For connected AI agents, Skovos can supply evidence mainly for elements 1, 3, and 4.
Sources
- Joint Commission and CHAI, Guidance on the Responsible Use of AI in Healthcare (RUAIH): digitalassets.jointcommission.org
- CHAI, AI governance workgroup: chai.org/workgroup/cross-cutting/ai-governance
- NIST, AI Risk Management Framework (AI 100-1): nvlpubs.nist.gov
- 45 CFR 92.210: ecfr.gov
- 45 CFR 170.315: ecfr.gov
- 45 CFR 164.504: ecfr.gov
- Health Sector Coordinating Council, Third-Party AI Risk and Supply Chain Transparency Guide, April 2026: healthsectorcouncil.org
- OWASP, LLM06:2025 Excessive Agency: genai.owasp.org
Related reading
Get a free AI governance assessment against the guidance elements
We review your AI inventory, owners, vendor agreements, and monitoring evidence against the seven RUAIH elements and give you a gap register. See a sample assessment.
Start free
A map of every AI agent in your health system.
We map who owns each agent, what it can access, and what evidence you could show your board or a surveyor.
Free. Asking takes 20 seconds. Please do not send patient information.