Show summary Hide summary
In a busy emergency room where decisions are measured in seconds, some clinicians on California’s Central Coast suddenly lost a digital tool they had begun to rely on. On April 6, Adventist Health barred staff from using publicly available conversational AI on hospital machines — a move that underscores how fast artificial intelligence is outpacing policy in health care.
Adventist Health’s ban and its immediate effects
Architectural Digest crowns Oklahoma river city the state’s most picturesque town
I-35 southbound crash kills one in train collision near SE 15th Street
The system-wide restriction applies to several Central Coast facilities, including Sierra Vista Regional Medical Center and Twin Cities Community Hospital. Doctors and nurses can no longer open consumer chatbots such as ChatGPT, Claude or Grok on hospital computers.
“The organization essentially hit pause until risks are better understood,” said Dr. Scott Bisheff, an emergency physician, describing the directive. For clinicians accustomed to consulting generative tools in clinical questions or workflow tasks, the change was abrupt.
Why some hospitals embrace internal AI while others step back
Use of AI among physicians has surged: a 2026 survey by the American Medical Association found that roughly 81% of nearly 1,700 doctors now incorporate AI in some form — a sharp rise from 2023. Most respondents reported that AI improved aspects of patient care, but adoption is uneven because of privacy and safety concerns.
At Dignity Health, for example, AI is active behind the scenes. The system runs checks on visits to catch overlooked findings. In one case a flagged small lung nodule detected on a chest X‑ray prompted a nurse practitioner to review whether the patient should be referred for cancer screening.
That approach differs from using open, public models because of patient privacy rules: under HIPAA, identifiable health details must be protected, and typing case specifics into a public chatbot could inadvertently expose protected information.
Health systems build closed models to keep data inside
CommonSpirit Health, which includes Dignity Health, has developed its own large language model, branded Insightli. The system is designed to be HIPAA‑compliant, to prevent patient data from leaving the organization, and to stop outside companies from using clinical inputs to train their models.
Dr. Monique Diaz, who helps oversee AI policy across CommonSpirit’s California hospitals, said the network evaluates new tools against three core questions: is it safe, is it private, and does it add clinical value? If a solution can’t meet those standards, it won’t be deployed, she said.
- Patient privacy: Closed systems aim to prevent protected health information from being shared with third-party models.
- Clinical safety: Hospitals require verification steps to catch AI errors before they influence care.
- Operational impact: AI can speed tasks (documentation, flagging findings) but also requires new workflows and oversight.
- Legal guardrails: California’s Physicians Make Decisions Act now prevents insurers from letting an AI system alone determine or alter a patient’s care plan; a clinician must make the final call.
Why institutions move cautiously
Large health systems tend to change slowly; their leaders compare them more to ocean liners than speedboats, said Robert Turbow, a neonatologist and Cal Poly professor who studies AI and patient safety. He warns that the technology’s pace can outstrip a hospital’s ability to catch errors.
Legal episodes in other sectors add to the caution. Courts and professional organizations have sanctioned lawyers for submitting briefs that contained fabricated citations generated by AI — a vivid example of how automated systems can invent facts. In medicine, even a single unchecked mistake can have grave consequences.
“Health care operates with very low tolerance for error,” Turbow said, noting that occasional AI falsehoods are unacceptable when patient outcomes are at stake.
Training and nuance: not a binary choice
Not everyone favors an outright ban. Ximena Escobar Greatorex, a master’s student researching AI risk in health care, argues against an all‑or‑nothing posture: clinicians need measured, evidence‑based integration and targeted training so tools are used safely.
Many proponents say the right balance is supervised use — limited, auditable, and always with human verification. For busy clinicians, that can mean quicker access to references and decision support if safeguards are in place.
On the front line: clinicians adapt
In the emergency department the work goes on. Bisheff says he treats some AI tools like an interactive, citation-backed reference that can save time — but he also accepts that decisions ultimately rest with human clinicians.
For now, the patchwork of bans, internal models and developing laws means health systems will keep refining policies as the technology evolves. The question for patients and providers is not whether AI will play a role, but how to harness it without compromising privacy or safety.
As hospitals navigate that balance, expect more institutions to formalize rules, invest in private models, or restrict consumer tools until clearer standards and oversight are in place — a transition that will determine how quickly AI becomes a routine, trusted part of clinical care.
