Research & data
Health System AI Governance Disclosure Index 2026
Of the 50 largest US health systems by hospital count, 34 (68%) publicly show at least one of six AI governance indicators in a source that was opened and quoted, and for 16 (32%) none was found on the public web. The count covers n = 50 systems, was reviewed on November 11, 2026, and counts sources that loaded and supported the indicator.
Results
Each system was scored on six indicators. The table counts systems where the indicator was found as Yes or Partial in a source that was opened and quoted. "Yes" counts those stated directly on the system's own page or release.
| Indicator | Yes or Partial | Percent of 50 | Yes |
|---|---|---|---|
| AI principles or responsible-AI statement | 13 | 26% | 6 |
| Named AI governance body | 26 | 52% | 4 |
| Named executive AI leader | 22 | 44% | 5 |
| Patient-facing disclosure of AI use | 7 | 14% | 1 |
| AI inventory, registry, or review process | 24 | 48% | 4 |
| Membership in an external AI standards group | 13 | 26% | 12 |
24 systems (48%) show three or more of the six. 1 shows all six at Yes or Partial, and none shows all six as Yes.
What stood out
- Governance bodies were the most common indicator. 26 of 50 (52%) have a committee, council or board that could be verified. Just 4 of those are stated directly on the system's own page. The rest come from trade press, interviews or statements that imply the body.
- Most evidence is Partial. Of 105 verified cells, 32 are Yes and 73 are Partial. Many indicators show up in trade coverage or interviews rather than on a system's own page.
- Executive AI leaders are named for 22 of 50 systems (44%). 5 are named on the system's own page or release.
- Review processes were found for 24 of 50 (48%). That means a public statement of an AI inventory, registry, or review or approval process. 4 are Yes.
- Patient-facing disclosure was the least common, at 7 of 50 (14%). 1 is Yes. Most of the others cover one tool or sit inside a privacy notice.
- External standards groups: 13 of 50 (26%). This is a floor, because the CHAI member roster is not public. 12 are Yes.
Examples where a system's own page states an indicator directly: Ascension describes a multidisciplinary AI Governance Committee. Ochsner says new AI tools are reviewed by its Data Governance and AI Steering committees. Kaiser Permanente publishes seven principles for responsible AI use. Banner Health announced a chief AI, data and infrastructure officer. Sutter Health names a chief AI officer on its site. These are examples, not a ranking.
Method
Who was counted. The ranking source is the AHRQ Compendium of U.S. Health Systems, 2023, sorted by hospital count. The top 50 were taken, with two substitutions. Steward Health Care System and Quorum Health Corporation were no longer operating as health systems, so they were replaced by the next two on the list, UNC Health and Appalachian Regional Healthcare. Names were updated to current names. Hospital counts are 2023 figures.
What was scored. Six indicators, each marked Yes, Partial, or No evidence found:
- AI principles or responsible-AI statement.
- A named AI governance body, such as a committee, council or board.
- A named executive AI leader, such as a chief AI officer or equivalent.
- A page or notice telling patients how AI is used in their care.
- An AI inventory, registry, or review or approval process.
- Public membership in CHAI, Health AI Partnership, TRAIN or a similar group.
Yes or Partial. Yes means the system's own page or official release states the indicator directly (for the standards indicator, the group's own page or release). Partial means credible third-party reporting, a job listing, or an own-site statement that implies it without stating it. Every Yes or Partial has one source URL and a quote of 25 words or fewer.
Checking. The research was run with AI research agents under my direction. Every source URL was fetched in November 2026 and a quote recorded. A seeded random sample of 30 verified cells (seed 20261111) was then re-fetched. 25 passed. 5 failed on whether the quote supported the indicator, so those were corrected and every cell for the systems involved was re-checked. The review date is November 11, 2026, after an initial pass on November 10.
Limitations
- This covers the public web only. No evidence found means a search of the public web did not turn it up. It does not mean a system lacks governance.
- CHAI's member roster is not public, so the standards indicator is a floor.
- 8 cells point to pages that blocked automated reading. They are excluded from every count here and from the data file.
- The Yes and Partial rule was applied by AI research agents and spot-audited, not by two independent human coders. Partial involves judgment.
- It is a point-in-time snapshot. The ranking uses 2023 data and hospital count, and systems have changed since.
Data, citation and corrections
The data is released under CC BY 4.0. Download the CSV: one row per verified finding, with the system, indicator, status, source URL and quote. A second file lists the 50 systems reviewed.
How to cite: Haver, W. (2026). Health System AI Governance Disclosure Index 2026. Skovos. https://www.skovos.ai/research/health-system-ai-governance-disclosure-2026/
Corrections: health systems can send corrections to contact@skovos.ai, and the data will be updated.
Disclosure: Skovos sells AI governance software to health systems.