Kaiser times compassion: the AI watching its nurses reveals the limit of managing care like a factory

By Zendoric · July 19, 2026.

If you call Kaiser Permanente's advice line because you have just received a devastating diagnosis, a licensed nurse answers. But that nurse may be watching the clock: exceeding 15 minutes on a call can end in a performance review meeting. That is how seven nurses, current and former, at Kaiser's call centers describe it in an investigation by CalMatters and The Markup published this month.

According to their accounts, software tracks the duration of each call and calculates monthly productivity scores. For several months in 2024, Kaiser also tested an artificial intelligence tool that assessed the tone and empathy in the voices of both nurses and patients. Raquel Alvarez Sánchez, a union delegate based in Vallejo since 2010, told CalMatters she had spent more than an hour on the phone with a suicidal patient while waiting for police, knowing that time would hurt her average for weeks. Another nurse, who requested anonymity for fear of retaliation, said she had held back from comforting an elderly woman in shock after a terminal diagnosis, afraid that lengthening the call would damage her score.

The breaks between calls have also been shortened: from about 10 minutes to complete notes or recover after a difficult conversation, to 30 seconds or less during peak hours, according to the nurses cited. The protocol also limits nurses to two or three pieces of advice per call. Kaiser denies the central premise: a spokesperson assured CalMatters that the company "does not use Average Handle Time to evaluate performance or to impose duration metrics" and that any AI tool in its contact centers has "human review and oversight." The company does not detail which systems it uses internally, citing operational security reasons. The tone-scoring tool was suspended in November 2024, though union representatives were warned it could be reactivated.

The context is not minor: Kaiser is California's largest private employer and serves more than 12 million people in the United States, according to CalMatters. The California Nurses Association is negotiating a new contract this summer for 25,000 nurses, about 1,000 of them in call centers, with AI as one of the central issues; there was already a one-day strike in March. In parallel, California lawmakers are considering rules that would protect doctors and nurses from retaliation if they override automated recommendations from an AI system. And Kaiser's own record adds weight to the complaint: the insurer paid a record $50 million fine for delaying behavioral health appointments and steering patients to group therapy instead of individual therapy, according to California's Department of Managed Health Care, in addition to a settlement with the federal Department of Labor over its mental health and addiction service.

Our reading is that this case is less about AI than about what an algorithmic management system is designed for. Here the technology was not used to free up nurses' time for care, but to compress that time and turn compassion into a variable that penalizes the monthly balance sheet. It is the AI application that fits least with the sector's underlying promise: when software measures and punishes the length of a conversation with someone in crisis, it is not augmenting human capacity, it is replacing it with a call-center metric applied to life-or-death decisions. The Cornell researcher cited, Virginia Dolleghast, sums it up well: the stress and burnout induced by surveillance raise the risk of error, and in healthcare that risk is paid for in lives, not in a bad review.

This also illustrates a tension we have been pointing out in the relationship between AI and employment: the healthcare sector is, in general, one of the most protected from automation because care and human relationships hold up better than routine tasks. But that shield breaks when management itself uses AI not to replace the task, but to squeeze the worker who performs it through surveillance and quotas. The underlying dispute —whether Average Handle Time is used to evaluate performance, as Kaiser denies— is exactly the kind of opacity that the next round of AI labor legislation should force into the open: what the system measures, what counts in the evaluation, and who has the final clinical say.

In the long run, we remain convinced that AI in healthcare can free doctors and nurses from administrative burden and help diagnose and treat better, allowing human time to be devoted to what no machine replicates: comfort, clinical judgment, presence in someone's worst moment. But that future requires AI to be deployed to augment those who care, not to time them. The Kaiser case is a short-term reminder that, without clear governance, the same technology that promises abundance can also be used to extract more from less, and that the union negotiation beginning this summer in California may end up setting the standard for how AI is deployed in healthcare across the country.