By James Whitaker

When a dialysis patient arrived at his unit, the AI-powered clinical support tool flagged a recommendation: load him with fluids. The nurse caught the error just in time. The patient had a condition that made fluid loading potentially fatal — a nuance no algorithm had been trained to flag, but one that a bedside nurse recognized in seconds, as documented in the dialysis patient fluid-loading near-miss reported in February 2026.

That near-miss has become a parable in an increasingly urgent debate: as artificial intelligence penetrates the most consequential corners of health care — triage, medication management, discharge planning, clinical documentation — the professionals with the deepest knowledge of real patient care are, with striking consistency, not in the room when these systems are designed, selected, or governed.

An investigation recently published, adds new investigative weight to that charge.

Hospital AI Governance Is Missing Nurses at Every Level

The numbers are stark. The ongoing Nursing AI Watch investigation, which tracks AI deployments across 106 large U.S. health systems, found that of those 106 systems, at least 26 have a named AI governance body. Of the governance structures that could be documented in detail, only six showed a named nurse leader publicly shaping how AI is governed, according to the nursing AI governance gap data published in July 2026. Just 16 of 106 systems publish a patient-facing AI disclosure policy, and only 12 of 106 have a confirmed Chief Nursing Informatics Officer.

That structural invisibility tracks with how nurses themselves report their experience. A global survey — Elsevier's Clinician of the Future 2026: Nurses Edition, drawing on responses from 692 nurses and 2,065 doctors across 118 countries — found that 41% of nurses say their views are rarely or never adequately represented in their organization's decision-making, compared to 19% of doctors who say the same about their own views, per the Elsevier nurses AI report. The usage gap is equally telling: only 41% of nurses report using AI at work, compared to 57% of doctors, and just 42% of nurses consider AI tools trustworthy.

Jan Herzhoff, President of Elsevier Health, said in remarks accompanying the report that realizing AI's potential "requires more than access — it demands trusted, evidence-based tools, proper training, and inclusive implementation that supports all members of the care team."

Meanwhile, AI-powered documentation tools, clinical decision support systems, scheduling algorithms, and patient monitoring platforms are being deployed across hospitals and health systems at a pace that is outrunning the ability of nurses to understand, evaluate, or influence them.

Why Bedside Knowledge Cannot Be Compiled Into an Algorithm

Nurses and nursing scholars argue the exclusion is not merely unfair — it is clinically dangerous. Bedside nurses catch what algorithms miss: subtle changes in affect, pain responses inconsistent with charted vitals, the social context that turns a routine discharge into a crisis. When AI tools are built and deployed without that expertise, they encode a partial view of care.

The technical reason is structural. Clinical decision support systems — the software category that underlies most hospital AI tools — work by matching a patient's documented clinical characteristics to a computerized knowledge base and generating a recommendation. The limitation is in what gets documented: the knowledge base is built from structured data, which typically means diagnoses, lab results, and medication orders. What nurses know at the bedside — the dialysis patient's contraindication that no code in the chart flagged — exists largely as unstructured, contextual clinical judgment that current clinical decision support architectures cannot reliably capture, as explained in the AHRQ clinical decision support primer.

A scoping review published in BMC Nursing in 2025 by researchers at Sorbonne Université and Lebanese Hospital Geitaoui-UMC found that clinical decision support development specifically built for the nursing process "remains in its infancy," with current systems lacking full coverage of the nursing process steps and the necessary linkages between them, per the BMC Nursing CDSS scoping review. Tools are frequently designed by and for physicians, then handed to nurses as finished products.

When training and governance skip the bedside, nurses end up using generalist chatbots for patient education and clinical questions — tools that do not always produce reliable answers for clinical use.

What Hospitals Are Doing to Nurses With AI, Not for Them

The frustration has moved from hospital corridors into union halls and onto picket lines.

At Montefiore Health System in the Bronx, the New York State Nurses Association alleged in July 2026 that Montefiore laid off 12 utilization review nurses — experienced clinicians who use decades of bedside judgment to review insurance denials and secure patient coverage — and replaced their work with AI-powered software from Datavant, a health IT company, per the NYSNA Montefiore nurses layoff press release. Datavant has two reported partnerships with Palantir, the data company widely used by U.S. Immigration and Customs Enforcement. Datavant previously agreed to pay $900,000 to settle a class-action lawsuit over a May 2024 phishing-related data breach, reported to federal regulators as affecting approximately 320,702 individuals — and the company denied wrongdoing, per the Datavant phishing breach settlement. Montefiore called the union's characterization of the layoffs "inaccurate and misleading."

In California, the conflict at Kaiser Permanente has grown into one of the most high-profile AI labor disputes in the country. A CalMatters investigation published July 9, 2026, documented that call-center nurses at the health giant are subject to call monitoring, predictive productivity software, and monthly performance scores based on speed and activity, per the CalMatters Kaiser nurses surveillance investigation. Nurses told CalMatters they are questioned about calls lasting more than 15 minutes and that software records active and inactive time, attempting to predict whether nurses are unproductive or answering too slowly.

Raquel Alvarez Sanchez, a Kaiser advice nurse in Vallejo and union steward, told CalMatters she spent more than an hour on the phone with a suicidal patient — knowing the entire time that it would distort her average call time for weeks and invite questions from management. Another Kaiser nurse told CalMatters that she chose not to offer comfort to an elderly terminally ill woman who was in shock after a cancer diagnosis, because she feared a performance reprimand. Nurses also described a voice-analysis AI Kaiser began testing in summer 2024 that attempted to assess empathy and tone in their calls with patients.

Kaiser says it does not use average handle time to assess nurse performance and that its contact-center tools support quality assurance subject to human review. The company did not explain to CalMatters why nurses are questioned about calls exceeding 15 minutes or why call time appears in the scores they described.

Nurses Ask: Should Algorithms Have More Authority Than a Nurse's License?

Faced with governance structures that exclude them, nurses have increasingly turned to collective bargaining to force their way in.

In New York, contracts ratified in early 2026 at systems including Mount Sinai, Montefiore, and NewYork-Presbyterian — after a 41-day strike by nearly 15,000 nurses that became the longest in a major New York hospital in years — included explicit technology protections: language specifying that AI cannot be used to replace nurses, to discipline them, or to drive staffing decisions, per the NYSNA AI contract ratification victory. In California, nurses with the National Nurses United and California Nurses Association secured AI language across the University of California's medical centers, and Sutter Davis Hospital's first-ever union contract guarantees nurses a say in how new technology like AI is implemented, per the AI contract bargaining tracker.

The use of AI in hospitals has become a friction point in labor-management disputes over the past year, and new contracts have begun integrating AI-specific language — though research found only seven of more than 100 tracked health systems have union contract language specifically addressing AI.

At the legislative level, California nurses have backed bills including AB 1883, which would prohibit employers from using AI to predict the emotional state of their employees, and AB 2575, which would protect health workers who override AI recommendations from retaliation, per the California AI healthcare worker bills. Federal policy has largely lagged. There is still no federal law setting AI rules for nursing practice, leaving a state-by-state landscape that shifts quickly.

Nursing associations Issue First Professional Consensus on AI Guardrails

In April, the nursing associations convened what it called its inaugural AI in Nursing Practice Think Tank — an invitation-only gathering of nursing leaders spanning practice, education, research, regulation, industry, and policy — at its Silver Spring, Maryland headquarters. The resulting consensus report, released May 5, 2026, identified five material risks already appearing in clinical workflows: erosion of professional judgment through overreliance on AI outputs, unclear accountability and liability when AI tools influence care decisions, algorithmic bias with the potential to exacerbate healthcare disparities, alert fatigue from poorly implemented technology, and the absence of nursing-specific governance standards, per the AI nursing practice press release.

A fundamental principle established in the report is that AI must support, not replace, professional nursing judgment — and that nurses remain the final accountable decision-makers. Nursing associations also called for mandatory AI literacy as a core professional competence for all registered nurses. Strategic priorities outlined, include curating a nursing AI playbook strategy and strengthening national advocacy for nurse-led AI governance.

Brad Goettl, DNP, DHA, APRN, Chief Nursing Officer of the American Nurses Enterprise, said the profession is "at a pivotal moment that requires deliberate, nurse-led action to protect patient safety and sustain public trust."

A 2026 paper in the Journal of the American Medical Informatics Association argued that nursing perspectives must be embedded in health system AI governance frameworks to ensure AI tools are clinically effective and trusted by the clinicians who use them daily, pointing to Duke Health's algorithm oversight framework as a practical model, per the JAMIA nursing AI governance framework.

Is This a Patient Safety Issue or a Labor Issue?

The governance gap is not merely a labor issue — it is a patient safety question with structural roots.

Nurses are, by most measures, the last line of defense in the clinical environment. They catch medication errors, observe deterioration before vitals shift, and interpret what a patient's family says in the hallway. AI systems that do not account for those functions — or that create pressure to override them in the name of throughput — introduce risk that no dashboard can fully capture.

No public evidence has yet definitively linked Kaiser's nurse surveillance to a specific patient injury, though nurses face a practical barrier: once they end a call, they may never learn what happened next. Failures in healthcare are frequently difficult to trace to a single decision, especially at the front door of the system. A patient who feels rushed may delay seeking care, as tracked by Kaiser AI surveillance tracking.

That traceability problem — the same one that makes it difficult to attribute harm to any individual AI recommendation — makes it all the more important that nurses are embedded in governance structures from the start, rather than brought in after the fact to troubleshoot tools already deployed. As the nurses investigation underscores, health systems are moving fast. The infrastructure of oversight — who sits on the committees, who reviews the models, who can escalate a concern about a failing tool — has not kept pace. And the people best positioned to spot the failures are, in hospital after hospital, still waiting to be asked.

Frequently Asked Questions

Are nurses involved in hospital AI governance decisions?

Rarely, according to current data. The Nursing AI Watch survey of 106 large U.S. health systems found that of the 31 governance structures it could document in detail, only six showed a named nurse leader publicly shaping how AI is governed, per nursing AI governance gap data. Of those 106 systems, just 12 have a confirmed Chief Nursing Informatics Officer. The 2026 consensus report named the absence of nursing-specific governance standards as one of five material risks already appearing in clinical workflows.

What happens when a clinical AI tool makes a dangerous recommendation?

In the most documented case to date, a nurse at a dialysis unit caught an AI recommendation to load the patient with fluids — a potentially fatal suggestion for a patient with a condition that contraindicated it. The nurse's bedside knowledge caught what the algorithm missed. This is the structural problem: clinical decision support systems are trained on structured clinical data (diagnoses, labs, medication orders), not on the unstructured contextual knowledge nurses carry from patient interactions. A 2025 BMC Nursing CDSS scoping review found that CDSS tools built specifically for the nursing process remain underdeveloped, lacking full coverage of nursing workflow steps. When hospitals deploy these systems without bedside nurse input into their design, recommendations can be wrong in ways the system cannot itself flag.

If a clinical AI recommendation harms a patient, who is legally responsible?

This is precisely the liability gap the consensus report named as a material risk: unclear accountability and liability when AI tools influence care decisions. Currently, no federal legal framework addresses this question for nursing specifically. Nurses are licensed practitioners of record, meaning they bear individual professional accountability for outcomes — yet they are routinely deploying clinical decision support systems they did not choose, were not trained on, and had no voice in governing. If a nurse follows an AI recommendation and a patient is harmed, it is not settled whether the nurse, the hospital, or the vendor bears legal exposure. Until federal standards create a clear answer, union contracts are the primary mechanism through which nurses are attempting to establish that AI cannot be used to discipline them or override their independent clinical judgment, per the AI in nursing consensus findings.

What AI protections do nursing contracts now include?

Contracts ratified in early 2026 after the 41-day New York City nurses' strike include three core protections: AI cannot replace a nurse, AI cannot be used to discipline a nurse, and AI cannot drive staffing decisions. New York's NYSNA secured this language at Mount Sinai, Montefiore, and NewYork-Presbyterian. In California, nurses won technology protections at University of California medical centers and Sutter Davis Hospital. However, Montefiore's July 2026 layoff of 12 utilization review nurses — which its union alleges violated the post-strike contract — shows that contract language is only as strong as its enforcement.