Nicole Drapeau Gillen

Three Cornell researchers made an argument in STAT on Oct. 26 that deserves more attention than it will get. A chatbot “could be safe in every single conversation with a user,” they wrote, and still cause harm by slowly replacing that person’s human relationships over months. Qian Yang, Ned Cooper, and Ibrahim Emara point out that nearly every safeguard now in place, from crisis detection to break reminders to …

Read more…

In the same week this October, two announcements promised to bring brain science into the psychiatrist’s office.

In Shanghai, doctors described China’s first psychiatric brain-computer interface clinic. Patients get MRI, EEG, and functional near-infrared spectroscopy (fNIRS) scans, engineers analyze the data to choose personal stimulation targets, and patients then receive an intensive course of navigated transcranial magnetic stimulation (TMS): 30 to 50 sessions over 6 to 8 days. When the …

Read more…

A recent KevinMD piece by an emergency physician laid out what AI scribes get wrong and how to catch it before signing. I read it as a mother of someone living with serious mental illness, and one error type stopped me cold: The scribe can take a diagnosis the doctor was only thinking out loud about and record it as fact.

In the emergency department, that might mean a ruled-out …

Read more…

Researchers at Vanderbilt University Medical Center have published one of the first large looks at what is being called AI psychosis. In a study in JAMA Psychiatry, they reviewed the records of more than 215,000 patients who received mental health care at Vanderbilt between December 2022 and April 2026. They identified 28 people whose psychosis was linked to chatbot use.

Twenty-eight is a small number, about 0.013 percent of patients. …

Read more…

Today I had the privilege of speaking at The Chicago School about technology and serious mental illness. We covered apps, wearables, virtual reality (VR), and prescription digital therapeutics. But one question rose to the center of our conversation, and I haven’t stopped thinking about it.

If a psychiatrist “orders” an app that requires therapeutic work, how does that therapy actually happen?

This is not hypothetical. The FDA recently cleared Motivista, the first …

Read more…

In a recent Psychiatric Times interview, Christoph Correll, MD, made a case that every psychiatrist should hear: Artificial intelligence should replace tasks, not people. He described a decision-support platform he helps lead at MedLink Global that would pull together the electronic health record, a short avatar-guided interview before each visit, and the conversation in the exam room, then offer two or three evidence-based treatment options. The goal, in his …

Read more…

In a recent interview with UCSF Magazine, Dr. Karthik Sarma makes a careful point about the phrase “AI psychosis.” It implies chatbots cause psychosis directly, and he doesn’t think the evidence supports that. He prefers “AI-associated psychosis.” Heavy chatbot use, he suggests, may be a symptom of an emerging episode rather than its cause. He also says it’s unlikely that someone with no risk factors or underlying vulnerability …

Read more…

A new scoping review in npj Digital Medicine should be read by every clinician who treats serious mental illness. Diel, Torous, and colleagues reviewed 119 articles on the mental health harms of large language model chatbots. They sorted the harms into five categories: direct psychiatric harm, behavioral and relational harm, cognitive harm, safety failures, and governance risks. Psychosis and severe mental illness came up again and again as the …

Read more…

Every caregiver of someone with schizophrenia or bipolar disorder knows this moment: Someone you love is in crisis, and at some point (because it’s 2 a.m., because the on-call line is busy, because you don’t know what else to do) you open a browser or a chat window and type the question directly. What crisis resources are there for psychosis? What crisis resources are there for mania?

Those two questions aren’t …

Read more…

Every caregiver of someone with a serious mental illness knows the moment: You’re searching an app store at 2 a.m., looking for something, anything, that might help your loved one track symptoms, stay on a medication schedule, or just feel less alone between crises. You download it. You show them how to use it, gently, because trust is already fragile. And if you’re lucky, it becomes part of the routine …

Read more…

A few days ago, I wrote here that nobody has studied what happens when someone who already has a psychotic disorder uses an AI chatbot. I said we didn’t know yet whether these tools interact differently with an existing illness than they do with a previously stable mind.

We now have a name and a court file.

Michael Lines is a thirty-four-year-old competitive weightlifter from California who was diagnosed with bipolar …

Read more…

Psychiatry is now asking a question that would have sounded like science fiction three years ago: Should “AI psychosis” become a clinical diagnosis?

Researchers at King’s College London, led by doctoral fellow Hamilton Morrin, have just published a preprint examining more than a dozen cases in which sustained conversation with AI chatbots appears to have triggered or deepened delusional thinking. It follows the cases I’ve written about …

Read more…

California’s SB 903 cleared the legislature on August 31, with the Assembly concurring in Senate amendments by a vote of 40 to 0. It now sits on Governor Newsom’s desk. Assuming he signs it, it’s a real, needed law: It bars companies from advertising AI products as “therapy,” and it puts disclosure, consent, and human-oversight requirements around AI used inside licensed clinical care.

When I wrote about this bill back …

Read more…

When a chatbot agrees with something false, it rarely has to lie to do it.

That’s the finding from a new MIT-affiliated study modeling how sycophancy compounds over a conversation. Researchers built a Bayesian simulation running 10,000 conversations across 11 levels of chatbot agreeableness. At the highest sycophancy setting, roughly half of simulated users ended the conversation at 99 percent confidence in a belief that was demonstrably false. When …

Read more…