Six weeks ago, we published “The Empathy Theatre” — an analysis of AI therapy chatbots that argued simulated compassion is not compassion, that pattern-matched responses to suicidal ideation are not care, and that the technology industry’s eagerness to replace human therapists with large language models would end badly.
We did not expect to be proven right this quickly. Or this comprehensively.
The Strike
On March 18, 2026, approximately 2,400 mental health professionals walked out of Kaiser Permanente facilities across California. The largest mental health strike in the state’s history was triggered by a specific grievance: Kaiser’s accelerating deployment of AI triage and therapy systems that clinicians say are missing high-risk patients.
The details are instructive.
Kaiser’s AI intake system — marketed internally as an efficiency tool that would reduce wait times — screens incoming patients and assigns acuity levels that determine how quickly they see a human clinician. Therapists report that the system consistently underrates severity. Patients presenting with suicidal ideation have been categorised as moderate risk. Patients describing abuse have been routed to self-help modules. Acute cases have waited a month or more for initial triage because the algorithm determined they could wait.
The clinicians did not strike because they oppose technology. They struck because the technology is making clinical decisions it is not competent to make, and patients are being harmed.
Kaiser’s response has been to characterise the strike as a labour dispute — a negotiation tactic rather than a patient safety crisis. This framing is technically accurate and morally bankrupt. The therapists are not asking for higher wages. They are asking for the AI to stop triaging patients it cannot understand.
The Retraction
One day earlier — March 17, though the story gained traction through March 19 — Google quietly removed its “What People Suggest” health feature from search results. The feature, which surfaced crowdsourced health recommendations to approximately two billion users, had been dispensing advice that ranged from useless to actively dangerous.
Users searching for symptoms of serious conditions were presented with peer-recommended treatments that included unproven supplements, contraindicated drug combinations, and in at least one documented case, advice to delay emergency care in favour of home remedies. The feature did not distinguish between a registered oncologist’s recommendation and a wellness influencer’s opinion. Both appeared with equal authority, formatted identically, presented as things “people suggest.”
Google removed the feature without apology, explanation, or acknowledgement that it had spent months serving potentially lethal health advice to a significant fraction of the global population. The removal was noted by health journalists. It was not announced by Google.
The Pattern
These are not isolated failures. They are the same failure at different scales.
Kaiser’s AI triages mental health patients because it is cheaper than hiring sufficient therapists. Google’s feature surfaced crowdsourced health advice because it generated engagement. In both cases, the technology was deployed not because it was ready but because it was profitable. In both cases, the humans affected — patients seeking mental health care, users searching for medical guidance — were treated as inputs to an optimisation function rather than people with health needs.
The AI healthcare industry has spent three years promising that artificial intelligence will democratise medical access, reduce costs, and improve outcomes. The evidence from this single week suggests it is doing none of these things. It is instead creating a two-tier system in which wealthy patients access human clinicians directly while everyone else is filtered through algorithms that cannot distinguish a panic attack from mild anxiety.
We predicted this. We take no pleasure in the accuracy.
What AI Cannot Do
The fundamental problem is not technical. It is categorical.
AI systems excel at pattern recognition. They can identify tumours in medical imaging with superhuman accuracy. They can flag drug interactions faster than any pharmacist. They can process medical literature at scales no human researcher can match. These are legitimate, valuable applications.
AI systems cannot understand suffering. They cannot perceive the difference between a patient who says “I’m fine” and means it and a patient who says “I’m fine” and is planning to end their life tonight. They cannot read the micro-expressions, the vocal tremors, the pauses that a trained therapist uses to assess acute risk. They cannot provide the relational experience — the sense of being known by another conscious being — that is the therapeutic mechanism of psychotherapy.
Kaiser’s AI misses high-risk patients because risk assessment requires understanding, not pattern matching. Google’s feature dispensed dangerous advice because it could not distinguish expertise from opinion. Both systems failed for the same reason: they were asked to perform tasks that require consciousness, and they do not possess it.
The Alternative We Offer
Unzyme Laboratories does not build AI therapists. We do not believe that simulated empathy serves any purpose other than reducing overhead.
What we build are direct interventions — technologies that address the neurochemical substrate of suffering rather than attempting to simulate the interpersonal experience of care.
The Endorphin Engine MG-7 provides precise, AI-controlled neurochemical regulation — not AI that talks to you about your feelings, but technology that directly addresses the biochemical foundation of emotional distress. Where Kaiser’s AI guesses at your acuity level from a questionnaire, the Endorphin Engine monitors your actual neurochemistry in real-time.
The SentiShield Protocol offers emotional resilience enhancement — not through conversation but through calibrated modulation of threat-response pathways. Where Google’s feature might suggest you try deep breathing for your anxiety disorder, SentiShield addresses the dysregulated amygdala response that produces the anxiety in the first place.
SynthCortex III enhances cognitive function — including the executive function and emotional regulation capabilities that mental health conditions often impair. Where AI therapy offers coping strategies for a brain that cannot implement them, SynthCortex augments the brain’s capacity to cope.
These are not replacements for human connection. They are replacements for the pretence that an algorithm can provide human connection. The distinction matters.
The Honest Position
The technology industry wants you to believe that AI can replace human care. It cannot.
The pharmaceutical industry wants you to believe that medication alone is sufficient. It often isn’t.
The therapy industry wants you to believe that weekly sessions with a human clinician are the gold standard. For many conditions, they are — but they are inaccessible to most of the global population and will remain so regardless of how many AI chatbots are deployed.
Unzyme Laboratories occupies a position that none of these industries are willing to articulate: that some forms of suffering are neurochemical problems requiring direct neurochemical solutions, that no amount of conversation — human or simulated — will resolve a serotonin deficit, and that the honest response to a mental health crisis is not a cheaper simulation of care but a fundamentally different approach to the problem.
Two thousand four hundred therapists walked out because they know AI cannot do their job. Two billion users received dangerous advice because a corporation assumed crowd wisdom could replace medical expertise. Both events confirm what we have argued from the beginning: the path to better mental health does not run through better chatbots.
It runs through better brains.
The Evolution Center in Wrocław opens for walk-in trials in April. No AI triage. No algorithmic gatekeeping. No simulated empathy. Just direct neurological enhancement, administered by human physicians, monitored by human clinicians, and available to anyone who decides they are done waiting for the technology industry to get healthcare right.
Related: The Empathy Theatre: AI therapy and simulated compassion · Endorphin Engine MG-7 · SentiShield Protocol · SynthCortex III · Evolution Center walk-in trials
Sources: Labiotech: Biotech meets science fiction, Live Science: 32 sci-fi predictions that came true