Nurses’ unions are now shaping how AI enters hospitals, through protests, member surveys, legislation and, increasingly, contract language that gives nurses a say before new technology is deployed. Their core demands are consultation, the right to override algorithms and protection of staffing levels. Managers who treat this as resistance to progress misread both the politics and the patient-safety evidence.
On 22 April 2024, hundreds of nurses from the California Nurses Association rallied outside Kaiser Permanente’s San Francisco medical center. Their placards read “Trust nurses, not AI.” Much of the coverage treated it as a novelty, nurses against robots, a scene from an older kind of labor dispute.
Two years later, it looks less like a novelty and more like the opening of a bargaining agenda. In 2025 nurses at Northwell’s South Shore University Hospital on Long Island won contract language allowing them to review AI technology before implementation, which the New York State Nurses Association described as a first for its contracts. In November 2025, nurses at Sutter Davis Hospital in California ratified a first contract that included a say in how new technology such as AI is implemented. In June 2026, nurses at Munson Medical Center in Traverse City, Michigan, voted 93% in favor of a contract that, among other things, defines AI and creates a formal route to raise concerns when new tools are introduced.
These are not bans. They are process rights. And the argument of this piece is that hospital leaders who ignore them, or fight them reflexively, are making a strategic error. Clinician unions are becoming one of the most important governance forces over clinical AI, and on several issues their demands align with what the evidence says good deployment looks like.
Why are nurses’ unions concerned about AI?
The simplest version of the answer is fear of replacement, but that is not what most union documents emphasize.
National Nurses United’s 2024 survey of more than 2,300 registered nurses is revealing. Sixty percent said they did not trust their employers to put patient safety first when implementing AI. About half said their hospitals used algorithms to determine patient acuity, and about two-thirds of those said the algorithm’s acuity did not match their own assessment. Twenty-nine percent said they could not override the system when they disagreed.
Acuity matters because it often drives staffing. An algorithm that underestimates how sick patients are can justify fewer nurses on a shift. So the complaint is specific: technology introduced without nurse input, feeding decisions that affect staffing and safety, with no way for the nurse at the bedside to correct it.
The survey is advocacy data from a union with a clear position, and it is not a representative sample of all nurses. But the concerns it records are concrete, and they are the same concerns that appear in union contract demands: consultation before deployment, the right to override, and a guarantee that AI will not be used to reduce staffing or replace professional judgment.
What are unions actually winning?
The contract gains so far are modest in wording and significant in principle.
The typical clause gives nurses a right to be informed and consulted before new technology is introduced, sometimes through a joint committee, and commits the employer to ensuring the technology supports rather than replaces nursing judgment. Union demands go further, seeking protection of staffing levels and an explicit right to override algorithmic outputs.
The New York State Nurses Association’s executive director, writing in the union’s magazine in 2025, framed the Northwell South Shore provision as a first for the union in New York and set it against broader concerns about virtual nursing and systems that monitor nurses’ work. National Nurses United has published its own “Nurses and Patients’ Bill of Rights” setting out principles for AI in care.
Alongside contracts, clinician organizations have pursued legislation. California’s SB 1120, the Physicians Make Decisions Act, sponsored by the California Medical Association and signed in September 2024, requires that decisions on medical necessity in health plan utilization review be made by a licensed clinician rather than an algorithm, effective January 2025. It targets insurers rather than hospitals, but it shows the same principle: humans must own consequential clinical decisions.
Why should hospital managers take this seriously?
There are at least four reasons, and only one of them is about industrial relations.
First, unions are right about some of the risks. The evidence on automation bias shows that clinicians follow incorrect algorithmic suggestions more often than they should, especially when the output appears before they form their own view and overrides are awkward. The right to override, and a process for reviewing disagreements, is good safety design, not just a labor demand.
Second, staffing evidence is strong. Research on nurse staffing, including a 2014 Lancet study across nine European countries, has linked higher patient-to-nurse ratios with higher surgical mortality. A union demand that AI not be used to raise workloads is also a patient-safety position.
Third, adoption depends on the people who use the tools. An AI deployed over the objections of the nursing workforce will be worked around, under-used or blamed for every problem. One deployed with nurses involved in selection, validation and workflow design has a much better chance of delivering what the business case promised.
Fourth, unions can make deployments slower and more expensive if they are ignored. In other industries, AI has become a strike issue. The Writers Guild of America’s 2023 strike in Hollywood ended with contract terms governing how studios could use AI. Health systems do not want AI to become the issue that turns a routine negotiation into a walkout.
A worked example: two ways to introduce virtual nursing
Consider a hypothetical unionized hospital planning to introduce virtual nursing, where experienced nurses monitor several rooms by camera and handle admissions, discharge education and documentation remotely, supported by AI tools that flag patients at risk of falls or deterioration.
Path one: management designs the model with a vendor, announces it, and presents it to the union as a done deal, alongside a revised staffing grid that assumes bedside nurses will take more patients because virtual nurses are handling paperwork. The union files grievances, members raise concerns about camera surveillance and alert accuracy, and the rollout stalls on the first two wards. Every adverse event during the pilot is blamed on the technology, fairly or not.
Path two: management raises the plan early with the union’s joint technology committee. Nurses help choose which tasks move to the virtual nurse, test the fall-risk alert in a silent trial on their own wards, and agree that bedside ratios will not change during the evaluation. The contract’s override clause is applied to the AI alerts from the start. The rollout takes three months longer. It also produces a model nurses support and data both sides trust.
The technology and the vendor are the same. The second path costs more time up front and much less later. It also leaves the hospital with something path one never produces: a group of frontline nurses who can explain the tool to colleagues, spot when it misbehaves and argue for it at the next negotiation instead of against it.
What is the strongest case against union involvement in AI?
The strongest objection is that unions may slow valuable innovation to protect jobs. Contract clauses requiring consultation can become effective vetoes. Some union rhetoric rejects AI in broad terms rather than engaging with the evidence for specific tools. And unionized staff are a minority in many health systems, so union positions may not reflect the wider workforce.
These concerns have some validity. Consultation that becomes indefinite delay helps no one, including patients who might benefit from a well-validated tool. And some union opposition is framed more categorically than the evidence supports.
But the objection assumes the alternative to union involvement is fast, safe innovation. The record suggests otherwise. Clinical AI has frequently been deployed without adequate local validation, without clinician input and without monitoring. Unions are forcing a governance step that many hospitals should have been taking anyway. The practical answer to delay is to build consultation into a clear process with time limits, not to avoid it.
What this means for hospital leaders, unions and policymakers
For hospital leaders, treat unions and frontline clinicians as part of AI governance, not as an obstacle to it. Put a nurse representative on the AI governance committee. Share validation data. Agree in advance what happens to staffing when a tool saves time. Our piece on the burnout number boards see argues that leaders often measure the wrong things about workforce strain, and AI deployment is a chance to measure the right ones.
For unions, the most effective position is specific: demand local validation, override rights, transparency about how outputs are used in staffing, and review of outcomes. Engaging with the evidence on individual tools makes union influence more durable than blanket opposition.
For policymakers, labor agreements are filling a gap left by thin regulation of how AI is deployed inside hospitals. Rules requiring clinician involvement, human accountability for consequential decisions and transparency about algorithmic staffing tools would reduce the need to fight each issue contract by contract.
For workforce planners, the context is a shortage, not a surplus. As we argued in our piece on nurse retention, nurses leave over workload and conditions. AI introduced in a way nurses experience as surveillance or staffing cuts will accelerate that leak.
Frequently asked questions
Why are nurses protesting AI?
Nurses’ unions have protested AI deployed without nurse input, algorithms that nurses cannot override, and concerns that AI will be used to cut staffing or monitor nurses. In April 2024, California Nurses Association members rallied outside Kaiser Permanente’s San Francisco hospital. Most demands focus on consultation, safety and staffing rather than banning technology outright.
Do union contracts include AI protections?
Increasingly, yes. Examples include a 2025 provision at Northwell South Shore University Hospital letting nurses review AI before implementation, Sutter Davis Hospital’s November 2025 first contract giving nurses a say in new technology, and a June 2026 Munson Medical Center contract that defines AI and sets a process for raising concerns.
What did the National Nurses United AI survey find?
The 2024 survey of more than 2,300 registered nurses found 60% did not trust employers to prioritize patient safety when implementing AI. Among nurses whose hospitals used algorithmic acuity tools, about two-thirds said the scores did not match their assessments, and 29% of respondents said they could not override the system.
What is California SB 1120?
SB 1120, the Physicians Make Decisions Act, was signed in September 2024 and took effect in January 2025. It requires that medical necessity decisions in California health plans’ utilization review be made by licensed clinicians, not by AI alone. It applies to insurers and was sponsored by the California Medical Association.
How should hospitals involve staff in AI decisions?
Hospitals can include frontline clinicians and union representatives on AI governance committees, share validation data, run silent trials on local wards, agree in advance how any time saved will be used, and create simple override and feedback processes. Early involvement usually shortens total implementation time by reducing later resistance.
Clinical AI will be governed by someone. If hospital leaders do not build that governance with their clinicians, their clinicians will build it for them, one contract at a time.
Has your organization discussed AI with its unions before a deployment, or only after the grievance arrived?
