Nurses using artificial intelligence tools at the bedside are reporting a specific and repeatable problem: the software tells them something their own assessment of the patient contradicts. In a National Nurses United survey of registered nurses, 69 percent of those whose employers use acuity algorithms said the results did not match their clinical judgment, and 60 percent said they did not trust their employer to put patient safety first when implementing AI. The survey drew responses from more than 2,300 nurses.
The objection is narrower than it sounds. Nurses are not rejecting technology. Their use of AI on the job nearly tripled in a year, climbing from 15 percent to 44 percent, according to Incredible Health's annual nursing report, a survey of 2,240 nurses that MedicalDaily covered in its reporting on a workforce adapting faster than its employers. What they are objecting to is being handed tools they had no part in selecting and no training to evaluate.
For patients, the governance question is not abstract. The nurse at the bedside is often the last check before an error reaches a person.
The Distinction Between Documentation Tools and Decision Support
Not all hospital AI carries the same risk, and lumping it together obscures where the safety argument actually applies.
Ambient documentation tools transcribe and summarize. They draft notes, populate fields, and reduce typing. When they fail, the failure is usually visible in the chart and correctable before it affects care, though independent research has found that errors in AI-generated notes are common enough that the human review step is doing real work.
Clinical decision support is different. These systems flag sepsis risk, generate a fall-risk score, push a deterioration warning to a screen, or calculate a patient's sickness level for staffing purposes. Their output enters the reasoning chain that determines what happens to a patient next. A wrong acuity score does not just produce a bad note. It can produce a nurse assignment that leaves too few staff on a unit with sicker patients than the algorithm recognized.
Administrative AI handling scheduling and billing sits further from the patient again. The 69 percent mismatch figure applies specifically to acuity measurement, the category closest to staffing decisions.
The Consensus Report That Named the Risks
The American Nurses Association convened its first AI in Nursing Practice Think Tank in Silver Spring, Maryland, in the spring and published consensus findings identifying risks already appearing in clinical workflows.
The ANA's consensus report names concerns about the erosion of professional judgment due to overreliance on AI outputs, unclear accountability and liability when AI influences care decisions, algorithmic bias that could worsen patient safety and health disparities, increased cognitive burden from poorly implemented tools, and the absence of nursing-specific governance standards. Its core principle is that AI must support rather than replace professional nursing judgment, and that nurses remain the accountable decision makers.
The liability item deserves attention from both patients and clinicians. When an algorithm contributes to a decision that harms someone, the question of who is answerable has not been settled. In a questionnaire of registered nurses conducted for Arkansas State University, 64 percent said they would not feel legally protected if an AI tool caused patient harm, and 35 percent said they felt pressured by their employer to use AI tools. That survey included 135 nurses, a small sample that limits how far the figures can be generalized.
National Nurses United has argued for a precautionary standard that places the burden on institutions to demonstrate a tool is safe and effective before it reaches the bedside.
Where Nurse Input Is Actually Being Won
The mechanism producing change is not voluntary consultation. It is contract negotiation.
Nursing AI tracking has identified seven health systems with ratified contract language addressing AI, covering tens of thousands of registered nurses, concentrated in New York and California. In New York, contracts ratified this year by systems including Mount Sinai, Montefiore, and NewYork-Presbyterian included explicit technology protections, including language stating that AI cannot be used to replace nurses, discipline them, or drive staffing decisions. In California, nurses secured AI language across the University of California medical centers.
There is also evidence that inclusion changes adoption. Eighty-one percent of nurses who helped choose AI tools reported using them, compared with 62 percent of nurses who were never consulted. Involvement is not only an ethical matter. It appears to shape whether the technology gets used at all.
Roughly 40 percent of nurses say they have no meaningful input into how tools are selected and deployed at their workplace, and only about one in five believes nurses have a real seat at the table.
Questions Patients and Families Can Reasonably Ask
Patients cannot audit a hospital's algorithm, but they retain the most reliable safety tool available: speaking up when something feels wrong.
If a patient or family member believes a condition is worsening and the response seems slower than expected, saying so directly to the nurse matters. Many hospitals also operate rapid response systems that families can activate, and asking on admission whether the hospital has one is a reasonable question.
Patients can ask whether an AI tool contributed to a recommendation and how the care team reached its conclusion. Clinicians are not obligated to accept an algorithm's output, and documenting the reasoning behind accepting or overriding a suggestion is exactly what professional bodies now advise nurses to do.
Nobody should decline recommended treatment because artificial intelligence was involved somewhere in the process. Well-validated tools catch deterioration that humans miss, and the argument from nursing organizations is about implementation and oversight rather than about whether the technology should exist.
The ANA has committed to issuing nurse-led guardrails and curating a nursing AI playbook, and state legislatures have taken up dozens of measures touching health AI, several addressing clinical scope and disclosure directly. MedicalDaily will report on those measures and on any federal action establishing validation or disclosure requirements for clinical algorithms.
Key Questions Answered
What are nurses objecting to? Not artificial intelligence itself. Surveys show nurses are adopting AI rapidly. The objection is to tools selected and deployed without frontline input, training or validation transparency.
What is an acuity algorithm? Software that estimates a patient's severity of illness, often used to guide nurse staffing assignments. In one union survey, 69 percent of nurses whose employers use them said the results did not match their own assessment.
Are all hospital AI tools equally risky? No. Documentation tools that draft notes carry different risk than clinical decision support that flags sepsis or deterioration, which enters the reasoning chain determining what happens to a patient.
What risks have professional bodies identified? The American Nurses Association named erosion of professional judgment, unclear accountability and liability, algorithmic bias, increased cognitive burden, and the absence of nursing-specific governance standards.
Do nurses have any formal role in these decisions? Increasingly through union contracts. Tracking has identified seven health systems with ratified contract language on AI, concentrated in New York and California. About 40 percent of nurses report no meaningful input.
What can patients do? Speak up if a condition seems to be worsening, ask whether the hospital has a rapid response system that families can activate, and ask how the care team reached their recommendation.
Should patients refuse care involving AI? No. Validated tools can catch deterioration that humans miss. The concern raised by nursing organizations is about implementation, oversight, and accountability.
