In the chaos of monitors and rolling gurneys, an unstable patient is rushed in and Dr. Scott Bisheff has only seconds to decide which tools to reach for.
One of those tools is now off-limits.
On April 6, Adventist Health’s board of directors restricted publicly available AI from internal use across its Central Coast hospitals. Physicians lost access to chatbots like ChatGPT, Claude and Grok on hospital computers.
“The corporation is saying, ‘Whoa, whoa, whoa, we’re not sure of the risk, so we’re not going to let you use it right now,'” said Dr. Scott Bisheff, an emergency medicine physician.
The policy applies only to Adventist Health facilities, including Sierra Vista Regional Medical Center and Twin Cities Community Hospital, and does not affect other Central Coast hospital systems.
A 2026 American Medical Association survey of approximately 1,700 doctors found that 81% are already using AI tools in their work — up from 38% in 2023. More than three-quarters say the technology improves their ability to care for patients.
The tools are moving faster than institutions can decide how they should be used – in clinics, in chatbots patients consult before appointments and in the background of hospital software.
On the Central Coast at Dignity Health hospitals, AI now reviews a patient’s visit to catch anything missed, said Dr. Monique Diaz, chief medical information officer for Dignity Health’s California region.
She gave one example: A patient comes in after a fall, a chest X-ray turns up a small lung nodule, and the system flags it to nurse practitioner Alexis Paulson. Paulson then checks whether the patient needs lung cancer screening.
AI is used across Dignity Health, but not the version anyone can pull up online. Under HIPAA, the federal health privacy law, almost any detail that could identify a patient is protected, and a single question typed into a public AI tool could give it away.
Therefore, CommonSpirit, the parent of Dignity Health, built its own large language model called Insightli. It is HIPAA-compliant and fenced off so that patient information never leaves the organization, is never used to train outside models and can be reviewed internally for misuse.
Diaz oversees AI policy across the system, including French Hospital Medical Center in San Luis Obispo, and helped shape how the technology would and would not be used across dozens of hospitals and clinics.
“AI is like a shiny object right now, where everyone wants to grab it and go after it, but we have to treat it like everything else,” Diaz said. “Is it safe? Is it private? Does it help? If you don’t pass those hurdles, you’re not going any further, even if you’re the shiniest object around.”
Ximena Escobar Greatorex, a master’s engineering management student at Cal Poly, does not speak for Adventis Health, but she has spent the past year researching AI risk in health care and built a curriculum on it for Western University of Health Sciences’ medical school.
“You can’t have the all-or-nothing mentality,” Escobar Greatorex said. “We can’t say never, because it’s here. But being super on board for everything also has its risks.”
Robert Turbow, a neonatologist and Cal Poly professor who teaches AI and patient safety, said the gap between what the technology can do and what hospitals are ready to use keeps widening, in part because large health systems are built to resist fast change.
“They have multiple layers. They’re ocean liners, not speedboats,” he said.
Turbow points to the legal world, where attorneys have been sanctioned for filing court briefs with case citations the AI simply invented. It is a warning, he said, about overreliance and skipping the step of verifying. The deeper fear is that AI will fail and no one will catch it.
“There’s an incredible amount of caution in the medical field,” Turbow said. “We work in a zero-tolerance zone for mistakes. If you’re using it and it’s occasionally wrong, it’s not good enough. That could be a matter of the patient not making it.”
Under a new state law, the Physicians Make Decisions Act, insurance companies can no longer let AI alone deny, delay or change a patient’s care. A doctor has to make that call. The California Medical Association, the state’s largest physician organization, says its goal is for the technology to strengthen medicine, not replace the people who practice it.
“I think of it as a reverse textbook,” Bisheff said. “Rather than looking up a question, you can just ask it and it’ll give you the answer with references.”
Back in the emergency room, the monitors still beep and the gurneys still roll. Bisheff makes his calls, whether the AI tools are within reach or off-limits.
“It’s the wild, wild West right now,” Turbow said, “which makes it kind of exciting, to be honest with you.”