The AI Psych Brief

Characterizing Artificial Intelligence Psychosis in a Large Medical Setting

Bergson Z, Vassall SG, Wright A, McCoy AB, Schafer KM, Achee MC, Sheffield JM · JAMA Psychiatry · October 2026

Vanderbilt investigators searched behavioral-health progress notes (December 2022 to April 2026) for AI-related keywords, kept only patients with a documented psychotic disorder, and had two raters sort each case into neutral chatbot use, "AI psychosis" (chatbot use that worsened psychosis), or AI appearing in delusional content without documented use. Across roughly 216,000 patients they found 28 cases of AI psychosis, about 0.013% of those in mental health care. Most were young (median age 27.5) and 60.7% were in their first psychotic episode, far more than in the comparison groups. The chatbot most often acted as an amplifier that validated existing distorted ideas (64%), less often as the object of the delusion (21%) or a catalyst for new symptoms (11%); 86% of cases were documented after the May 2024 GPT-4o release.

The design is retrospective and single-site, depends on what clinicians happened to write down, and cannot establish causation.

Read the paper →  doi:10.1001/jamapsychiatry.2026.2923

Shedding light on the efficacy of psychodynamic treatment targeting depression for patients with comorbid narcissistic personality disorder

Mansour M, Kivity Y, Zilcha-Mano S · Psychotherapy Research · October 2026

This secondary analysis drew on a randomized trial of 100 adults with major depression who received short-term supportive or supportive-expressive (psychodynamic) therapy; 14 met SCID criteria for comorbid narcissistic personality disorder. Depression was tracked on the Hamilton scale across treatment with multilevel models. Patients with comorbid NPD showed greater depressive symptom reduction than those without it. Their NPD diagnostic status, however, did not change significantly over the course of therapy.

The NPD subgroup is small, so the finding is a signal rather than a settled moderator effect.

Read the paper →  doi:10.1080/10503307.2026.2738631

Thiazides raise lithium within days; ACE inhibitors can take weeks

Thiazides cause mild volume contraction, and the proximal tubule responds by reabsorbing more sodium. It cannot tell sodium from lithium, so lithium is reabsorbed too. Steady-state levels typically rise 25 to 40 percent, and the change shows up within about a week. If a thiazide is unavoidable, reduce the lithium dose pre-emptively and recheck the level in five to seven days. Loop diuretics act downstream of where most lithium is reabsorbed, so they cause smaller and less predictable rises. They still need a level check, mainly because of dehydration.

ACE inhibitors and ARBs are the slower trap. They lower GFR and aldosterone-driven sodium retention gradually, so toxicity can appear three to five weeks after starting, often in an older patient whose first level came back normal. NSAIDs vary widely: indomethacin and diclofenac raise levels substantially, while low-dose aspirin and sulindac matter little. One normal level after any new antihypertensive does not end the monitoring.

A silence has a transference before it has a meaning

Residents tend to treat silence as a pause, either a gap to fill or a technique to hold. Dynamically, silence is a relational event, and it carries a transference before it carries any particular content. The useful question is not "what is the patient not saying?" but "who am I to the patient while we are silent?" The same quiet minute can be felt as an attentive parent's presence, a withholding judge's scrutiny, or an abandonment. Which one it is decides whether the silence helps or harms.

For a patient with a stable enough internal object, a held silence lets associations come forward. For a patient who experiences you as critical or absent, an unaddressed silence repeats the trauma instead of opening it. The technical move is to read the countertransference pressure first (do you feel waited on, shut out, or tested?), then address the experience of the silence before asking what it contains.