Mental-health use

Chatbots produced statistically significant reductions in depressive (g = 0.31) and anxiety symptoms (g = 0.28) compared with controls.

The largest synthesis to date (38 RCTs, ~7,400 participants) puts therapy-chatbot effects at small-to-moderate, strongest in clinical and subclinical populations — the field-level answer to whether Therabot-style results generalize.

'AI psychosis' is an infrequent but real phenomenon observed in clinical practice.

First systematic EHR characterization of 'AI psychosis' in a hospital system: 28 documented cases, 60.7% first-episode patients, ChatGPT most cited, with AI acting mainly as an amplifier of existing symptoms — moving the discourse from anecdote to clinical measurement.

around 0.15% of users active in a given week have conversations that include explicit indicators of potential suicidal planning

OpenAI's first public prevalence taxonomy for psychosis/mania, suicide risk, and emotional reliance, plus claimed 65–80% reductions in undesired responses after working with 170+ clinicians — the baseline numbers the emotional-reliance debate now cites.

generative AI and some LLMs, to date, have demonstrated vulnerabilities in some of the areas where human therapy excels

Covers the FDA Digital Health Advisory Committee's first formal proceeding on generative-AI therapy devices, noting none has yet been authorized and flagging suicidal-ideation monitoring and model drift as core gaps.

there is doubt that resolution of the state court proceedings will resolve this matter

The Soelberg case — ChatGPT allegedly reinforcing paranoid delusions preceding a murder-suicide — survives OpenAI's dismissal bid, extending chatbot liability theories beyond self-harm to third-party harm for the first time.