Some hospitals ask patients to sign a consent form before recording a physician’s visit. They don’t ask before an algorithm reads the patient’s EKG. That inconsistency is at the heart of a fight health leaders still haven’t settled: When, exactly, does a hospital owe a patient the truth about the machine in the room?
The question split a Sept. 23 panel on AI and clinical decision-making at Mass General Brigham’s World Medical Innovation Forum in Boston, moderated by Bret Bostwick, MD, a physician-turned-venture capitalist who asked his panelists directly: full disclosure, or does that risk creating more confusion and anxiety than it resolves?
Rebecca Mishuris, MD, vice president and chief health information officer of Somerville, Mass.-based Mass General Brigham, said the EKG is exactly the kind of case most people don’t think about. “We don’t disclose everything today, actually,” she said. “If you get an EKG, the top of the readout is an AI algorithm. We don’t tell patients that. In fact, I suspect that most clinicians don’t know that either.” Clinicians are trained to read the EKG themselves rather than rely on the automated interpretation, she said, but the AI reading is there regardless — and there’s no practical way to ask a patient’s permission first. “If a patient said, ‘I don’t want you to use that,’ what would we do? We wouldn’t be able to do an EKG. We wouldn’t be able to treat your chest pain.”
She contrasted that with ambient documentation tools, where clinicians do ask patients to consent before a visit — but not to the part most people assume. “You’re actually not getting permission to use an artificial intelligence model,” she said. “You’re getting permission to record the visit.”
Earlier this year, Rhode Island became the first state to require providers to notify patients when they’re using an AI scribe, while 12 states — including Massachusetts — mandate two-party consent for audio recording.
Not everyone on the panel saw as much nuance. “You have to disclose today. So right now you don’t have a choice, and I think it’s great to disclose,” said Allon Bloch, co-founder and CEO of healthcare AI startup K Health, noting that AI is already used extensively in hospitals, in radiology and elsewhere, often without patients realizing how much. “I don’t think there’s a way around not telling people that this is a conversation with a machine, as intelligent as it is.”
Dr. Mishuris pushed back on the idea that the answer is that simple. “I think the reason you struggle with it is because there is no right answer, right?” she said. “And really smart people disagree about whether we should disclose everything, whether we should disclose nothing.” She argued the debate needs to start further back than most people begin it: “I think if we go back to first principles, what’s the purpose of disclosure — that is where we need to start.” Is the goal to preserve a patient’s right to refuse? To give them grounds for a second opinion? Or is it about liability, and making sure the health system can show it was transparent? “If you start with what are we trying to achieve through disclosure,” she said, “I think you will get to as reasonable an answer as you can get to.”
“There is no single answer,” Dr. Mishuris said. “It’s not disclose everything or disclose nothing. It’s going to be based on the workflow, the technology and, really importantly, the purpose.”
Haider Warraich, MD, a cardiologist and program manager at the federal Advanced Research Projects Agency for Health who is investing heavily in agentic AI, argued later in the same panel that the disclosure debate may already be overtaken by events: Patients are turning to AI on their own, whether or not a hospital ever brings it up. He described a patient of his who had a stroke while running an errand. “She couldn’t speak,” he said. “And the first thing she went to was ChatGPT, and ChatGPT told her to go to the emergency room, and she did.” He said the major AI labs are likely undercounting how many people already use their tools for health questions — a trend he expects “is going to get smaller and smaller” in reverse, meaning fewer and fewer patients will be AI holdouts.
