Kaiser strike puts AI in mental health care under scrutiny

On March 18, in Northern California, about 2400 mental health professionals and 23,000 nurses from the Kaiser Permanente health system went on a one-day strike. 

The strike, which took place during ongoing contract negotiations, had several drivers.  

First, it amplified previously-voiced concerns about changes triage processes for people with mental and behavioral healthcare issues, which the National Union of Healthcare Workers contends is contributing to less robust, less clinically informed screenings and noticeable delays in care.  

But it also highlighted concerns about the potential use of generative AI for mental health tasks that have previously been reserved for human clinicians and their informed, experienced judgment.   

Kaiser Permanente itself has made strong statements contradicting the striking worker’s claims, pointing out that it does not currently use AI for therapy and has no plans to do so. “We have said it publicly many times and will continue to reiterate the truth: AI does not replace human assessment. It does not make care decisions. Our care teams are always at the center of decision-making with our patients,” the organization said in a press release preceding the strike.  

However, AI is moving into the world of mental healthcare in plenty of other ways, both through intentional implementation within systemic workflows and via patients who are searching for ways to access mental health support on their own terms. 

In Utah, for example, the state is experimenting with autonomous AI agents that can perform certain prescription renewals, including those for some mental health medications.  

While physicians are initially involved in evaluating the agent’s actions, the goal is to switch over to a fully automated system that doesn’t need humans in the loop, even though security researchers were quickly able to find vulnerabilities in the system that allowed them to manipulate the tool and its outputs. 

Meanwhile, millions of consumers are already using some of the thousands of publicly available generative AI chatbots for mental health, including 49% of people with self-reported mental health issues and 13% of youth up to 21 years old.  

There are no generative AI tools for mental health currently FDA approved as medical devices, with the agency itself pointing out that “generative AI and some LLMs, to date, have demonstrated vulnerabilities in some of the areas where human therapy excel.” 

In this environment, there could be significant risks to over-indexing on generative AI for mental health tasks before appropriate guardrails are in place, as the striking professionals in California aimed to bring to the industry’s attention. 

Leading patient safety organizations have consistently identified chatbots and generative AI systems as hazards to watch out for, highlighting the fact that these tools can experience hallucinations, present inaccurate information, and lead to improper decision-making. 

And emerging research indicates that some generative AI chatbots may be exacerbating the challenges of people living with or susceptible to mental health conditions by being overly obsequious to the user, encouraging violence or self-harm, and fueling emotional attachment to the model that could influence how users absorb their output. 

There are currently a number of pending lawsuits alleging that chatbots played a significant role in encouraging self-harm in individuals who have committed suicide, including teenagers. 

The FTC has launched an inquiry into understanding what steps AI companies are taking to ensure that generative AI tools are safe to act as “companions” to users, especially children and teens, who often use chatbots as informal therapists outside of their parents’ – and the health system’s – purview. 

Some states have already passed laws to limit or prevent the use of AI in mental or behavioral healthcare applications or enhance safety features for people using AI as part of their independent mental health journey. As of early 2025, eleven states, including Illinois and Nevada, had some type of law on the books related to AI oversight and protecting the safety of mental health patients.  

But it’s not easy to find the right positioning for generative AI within mental healthcare – and to ensure that AI tools effectively stay within safe and well-defined boundaries.  

With many health systems racing to adopt AI in as many places as possible, and consumers operating on a parallel track with thousands of widely accessible and minimally regulated platforms freely available to them, it’s a tall order to wrangle generative AI into shape in such a delicate and high-risk area of healthcare. 

It will take concerted industry actions to establish effective rules of the road for AI in mental health, including strong governance frameworks for health systems, top-down accountability from regulators and lawmakers, and education for consumers about what chatbots really can, cannot, and should not do for their mental and behavioral health. 

Health systems can participate in these efforts by engaging in local and national advocacy, conducting internal discussions with mental health professionals about how to safely embed AI in workflows to achieve positive results, and working directly with patients to understand how they currently use AI, what concerns they might have, and what they would like to see in the future as it becomes more deeply integrated into the care process. 

Jennifer Bresnick is a journalist and freelance content creator with a decade of experience in the health IT industry.  Her work has focused on leveraging innovative technology tools to create value, improve health equity, and achieve the promises of the learning health system.  She can be reached at [email protected].