В США критикуют внедрение ИИ в психиатрии за рост ошибок и отказ пациентов от лечения

This article was produced by Capital & Main, an award-winning publication that reports from California on economic, political, and social issues. It is co-published here with permission. When Kaiser Permanente triage clinician Harimandir Khalsa began working in the psychiatry department at Kaiser’s Walnut Creek Medical Center in Northern California, she was on a team of nine people. Today, just over three years later, she is one of only three triage clinicians left. Some of the work once handled by employees has shifted to automated and algorithmic tools. This change has coincided with a sharp increase in the number of patients who are upset by the time they speak to her. On a typical day, as many as a third of her almost two dozen triage calls are with patients who have struggled to access appropriate care. “What has increased is frustration. Sometimes people are like, ‘I’ve been sent all over the place,’” she said. The decline in triage staffing and the increasing use of artificial intelligence-powered systems at Kaiser have also led to dangerous delays, missed diagnoses and inappropriate treatment decisions, according to multiple Kaiser mental health care workers. To understand some of the ways that AI is affecting mental health care, Capital & Main spoke with more than a dozen therapists, clinicians, academic experts and advocates in Wisconsin and California, who said the hasty integration of the technology is eroding patients’ access to timely and appropriate mental health care. While AI is changing work in many professions, its use can have particularly high stakes in the mental health field, workers said, by creating barriers to care for patients who are already struggling with major depression, bipolar disorder and other serious conditions. Increasing transparency around its use and blocking the deployment of AI in triage decisions have become major goals in collective bargaining for these workers. Already something of a national leader in AI regulation, the state of California, where Kaiser is headquartered, now has pending legislation that would create safeguards around the technology in medical settings. Capital & Main interviewed mental health care workers employed by Kaiser and Rogers Behavioral Health, a Wisconsin-based provider of addiction and mental health treatment. Spokespeople for these health care providers denied that their AI systems are replacing the work of mental health professionals. “Clinical assessments, triage and treatment decisions are made by licensed clinicians, and any technology we use is designed to support—not replace—clinical judgment,” Kathleen Chambers, a Kaiser spokesperson, wrote in an email. Distressed Patients and Dangerous Delays About two years ago, Ilana Marcucci-Morris, a Kaiser therapist who works in Oakland, noticed that the path to mental health care was becoming increasingly dictated by technology. Instead of being referred directly by primary care doctors and contacted to schedule appointments, patients were routed through an app, nonclinical call center staff or simply given a phone number to call. “Now the onus is on the patients to be their own care coordinators,” she said. This dynamic can be particularly challenging for some patients. “Lack of motivation and lack of follow-through are the most common symptoms of depression on the planet,” Marcucci-Morris said. “We wouldn’t tell a paraplegic, ‘Hey, walk down the hall in order to get your wheelchair.’” Khalsa and Marcucci-Morris are now spending more time doing what is called “service recovery”—the effort to repair trust in patients whose expectations of care haven’t been met. Even when broader use of automated and algorithmic systems does not eliminate human jobs, it can change their nature, increasing the emotional labor needed to reassure patients disserved by those very systems. “More often than not, I’m de-escalating an upset patient because they were sent down the wrong path and are suffering, so they’re frustrated with us,” Marcucci-Morris said. Melissa Stevens, a clinical psychologist at Kaiser in Northern California who works in a chronic-pain program, will often make referrals within Kaiser when patients need psychiatric care for intense depression or delusions. Over the past year, however, she said that Kaiser’s reliance on an AI-powered triage system has made those referrals much harder. She sometimes hears from patients who were never contacted or who were placed in group classes for anxiety or depression, even when they needed a higher level of care. “It’s getting dismissed by whatever their filtering system is until I actually pick up a phone and call a human,” she said, adding, “I’m like, ‘What’s happening here?’” Such delays can be not only frustrating but also dangerous, said Khalsa. “I’ve definitely seen those cases where there’s been a delay of sometimes weeks with this new triage system or where self-harming or a prior suicide attempt has been missed,” she said. Kaiser now connects patients with multiple third-party apps and digital programs, such as Calm, which offers guided meditations, and Meru Health, a 12-week mental health program that includes up to four telehealth sessions with a licensed therapist and contains short videos and activities in its app. “Kaiser is saying, ‘Hey, it’s so great; you can just instantly get some cognitive behavioral techniques about sleep,’” Khalsa said. “But what if this person is in the midst of a manic episode, and they’re not sleeping because they’re manic? A little app isn’t going to help.” While Kaiser workers said these apps and programs can provide valuable information and insights in some cases, they told Capital & Main that in other cases they do little for patients, who get worse without appropriate treatment. “It’s not weekly therapy, and so I’ve had a lot of patients come back from that, and then three or four weeks in, they’re in worse shape than when they started,” Marcucci-Morris said. “The ones that really trouble me go through the full 12-week program, and then they go to a medical appointment again, and their depression is worse.” Inappropriate Placements In late 2024, Rogers Behavioral Health, which has facilities in 10 states, announced a partnership with the company Limbic, which offers AI products to help mental health care providers with intakes, case management, clinical assessments and other tasks. The partnership’s stated goal, according to a Rogers press release, was “to create easier pathways to care for anyone making the courageous decision to seek mental health support.” By August 2025, Limbic published a case study on Rogers’ use of its products, including an AI intake chatbot and an AI voice agent. The report claimed that Limbic used “real-time routing and prioritization based on AI-generated clinical intelligence.” For Rogers workers interviewed by Capital & Main, the integration of Limbic was not the roaring success that the report suggested. Erin Quinlan, a behavioral specialist at a Rogers facility in Madison, Wisconsin, has noticed many patients placed in an inappropriate level or type of care since Rogers began using Limbic. A recently admitted patient who was “a very severe suicidal risk” requiring inpatient care was instead admitted into “the lowest level of care” provided in her building: intensive outpatient care, Quinlan said. She estimated that in her behavioral health group classes, which usually have at least eight participants, typically at least one person should not have been placed in the class. “If a patient does not have the ability to regulate their emotions through skills, they will become overwhelmed incredibly quickly, and we will have them in crisis,” she said, noting that such crises have involved patients crying, screaming and attempting to leave the facility. In an email to Capital & Main, a Rogers Behavioral Health spokesperson, who asked that their name not be used, credited AI with “allowing our team members to spend more time focused on patient care,” adding that the tool was used to gather information from patients in “a HIPAA-compliant way.” The email also said that “only licensed Rogers clinicians make admission, placement and treatment decisions.” Limbic did not respond to multiple requests for a comment. Kate Zolandz, a former Rogers therapist, said she also saw inappropriate placements increase after the introduction of Limbic. Zolandz, who worked for four years in the company’s West Allis, Wisconsin, facility, was particularly struck by an apparent lack of screening for past aggressive behavior after the adoption of Limbic. “When they’re crying, they’re dysregulated, they’re screaming, they’re throwing things, it disrupts patient care, but it also stretches staff even thinner because they’re needing to attend to this immediate crisis,” she said. A Way Forward: Guardrails and the Uniquely Human Some of the problems identified by mental health care workers could in theory be improved with better AI. Yet experts and clinicians caution that some problems are only fixable by ensuring a central place for licensed human caregivers in mental health and other health care. Labor unions and legislators are currently proposing ways to place guardrails around the use of AI in union contracts and state law. “Ultimately, AI tools are just that: They’re tools. The provider is the trained licensed professional, and so their judgment is what’s important,” said Leanna Fortunato, a clinical psychologist and the director of digital health and innovation at the American Psychological Association. A key provision of California Assembly Bill 2575, authored by Assemblymember Liz Ortega, seeks to defend the judgment of professionals by protecting health care workers from retaliation if they override the recommendation made by AI. The bill would also require health care facilities to be more transparent with workers about the risks and uses of AI systems and would shift liability to AI developers and facilities for any harm to patients. The bill passed the Assembly by a wide margin and has since advanced through several Senate committees, where it awaits a floor vote. The California Hospital Association and Kaiser are among the signatories to a March letter opposing the bill. David Simon, senior vice president of communications at the California Hospital Association, wrote to Capital & Main, “At the heart of our concern is that this bill would undermine the many ways that AI utilized by clinicians through clinical decision support systems can improve nearly every aspect of health care—from quality, patient experience and affordability to clinician efficiency and well-being.” Robert Wachter, chair of the department of medicine at the University of California, San Francisco, and author of the 2026 book A Giant Leap: How AI Is Transforming Healthcare and What That Means for Our Future, was also critical of the bill. “We need some room for safe experimentation with oversight, as opposed to locking in the status quo,” he said. Ortega told Capital & Main she recognizes the importance of AI in improving health care and credited it with “saving lives and being able to identify serious health conditions.” However, she said, AI “can also get it wrong” and that patients and workers deserve protection and transparency from health care providers. Another pending bill, AB 1979, authored by Assemblymember Mia Bonta, would outlaw clinical decisions based solely on AI output and strengthen data privacy protections. The National Union of Healthcare Workers, National Nurses United and the California Nurses Association support both bills. (Disclosure: NUHW and CNA are financial supporters of Capital & Main.) In Northern California, Kaiser health care workers have made the appropriate deployment of AI a major issue in contract talks. However, Kaiser is currently refusing to agree to language stating that artificial intelligence “is not to replace but to assist” employees “in providing safe therapeutic and effective patient care and support,” according to NUHW. Chambers, the Kaiser spokesperson, did not respond to a question about the proposed language but said in a statement to Capital & Main: “As part of our ongoing negotiations with NUHW, our proposals are focused on ensuring our members have timely access to high-quality mental health care, supporting our clinicians in delivering excellent care and meeting the growing demand for services.” The roughly 2,400 Kaiser mental health care workers in Northern California represented by the NUHW have been without a contract since last September, and the health care giant’s hospital system’s use of AI has emerged as a major source of disagreement. That disagreement also prompted the NUHW to file a complaint with California’s Department of Managed Health Care in April last year, alleging that Kaiser uses AI to triage patients in violation of California law. The union also filed a second complaint on July 20, alleging that Kaiser’s web-based “e-visit” tool also violates state law. In written materials quoted by NUHW, Kaiser said its intake system uses “built-in logic and algorithm” to identify a patient’s level of distress before directing call-center agents to schedule care. The union is arguing that triage must legally be performed by licensed clinicians, a view shared by Robin Feldman, a professor of law at UC Law San Francisco and the director of its Center for Innovation. “California law requires that a licensed health care professional must make the decision about the level of care a particular patient needs,” Feldman wrote in an email. In February, Kaiser reached an agreement with the U.S. Department of Labor to pay more than $28 million to its patients who had to go out of network for mental health care between 2021 and 2024. The settlement announcement said that Kaiser “used patient responses to questionnaires to improperly prevent patients from receiving care.” Chambers said that this matter did not involve AI. Kaiser has claimed that a shortage of mental health care workers was partly responsible for its struggles to provide adequate care. Some of its workers cite Kaiser’s roughly $67 billion in unrestricted cash reserves to question that claim. Meanwhile, in West Allis, Wisconsin, Rogers’ mental health care workers voted 54-4 to join the NUHW in April, but Rogers has not yet agreed to meet with them and negotiate. Transparency on exactly how AI is used was a key issue identified by several workers who spoke with Capital & Main. Even with better AI systems and more transparency on how they are used, therapists at Kaiser and Rogers stressed that such tools cannot produce the same benefits as a strong relationship with a human caregiver. “Therapy works because you build a relationship with your therapist,” said Zolandz, the former therapist at Rogers. “AI can’t build that connection in the same way that an actual human being can.”