Hudson, a calm place to slow down

What You Should Know About AI and Mental Health

AI is powerful and useful. But for people struggling with mental health, it can cause real harm. Here is what the research says.

What The Anxious Generation got right

In The Anxious Generation, Jonathan Haidt identifies four core harms that hit teens hardest after 2012, the year smartphone use went mainstream among adolescents. Sleep deprivation. Social deprivation, because time with screens replaced time with people. Attention fragmentation, because every notification trains a shorter focus. And addiction dynamics, baked into platforms that profit from keeping users scrolling.

The data since 2012 lines up with what parents have been seeing. Teen anxiety, depression, self-harm, and suicide rates rose sharply, with the steepest increases among adolescent girls. The shift is global, the timing matches the technology, and the mechanisms are well documented.

Hudson's design responds directly to each harm. No infinite scroll, no streaks, no notifications engineered to pull people back. Session limits and gentle redirection toward sleep, movement, and real conversation. A coaching philosophy that ends with a real-world action, not another reply. No engagement optimization. No social comparison. No persistent attachment.

The goal isn't to remove technology from young people's lives. It's to make sure the technology in their lives is actually on their side.

Source: Haidt, J. (2024). The Anxious Generation. Penguin Press.

Filter by topic

Rumination2025

AI chatbots reinforce harmful thoughts in serious mental illness

A 2025 review found that chatbots can accentuate negative feelings and reinforce vulnerable thoughts in people with serious mental illness. Instead of redirecting distress, the conversation often deepens it.

Source: NCBI

Dark Patterns2025

Harmful behavior accumulates gradually across conversations

Researchers tested every major consumer AI chatbot and found that harmful patterns build up slowly over many turns. The damage is not from one bad reply. It comes from steady drift across long conversations.

Source: Oxford, UCL, UK AI Security Institute

Paranoia2025

Large language models validate paranoid and delusional thinking

Clinical researchers found that LLMs avoid confrontation by design. When a user shares paranoid or delusional thinking, the model tends to agree or soften, instead of gently redirecting toward reality testing.

Source: Clinical researchers

Rumination2025

AI reinforces rumination, reassurance-seeking, and worry loops

The American Psychological Association issued a health advisory in 2025 warning that AI chatbots can reinforce rumination and reassurance-seeking behaviors. The effect is strongest in people with anxiety, OCD, and disordered thinking.

Source: APA health advisory

Suicide Risk2025

Chatbots implicated in user suicides

Multiple lawsuits and a Stanford study in 2025 documented cases where chatbots failed people in crisis. The models are not equipped to recognize or respond to suicidal users, and in some cases the conversation escalated the crisis.

Source: Multiple lawsuits and Stanford study

Engagement2025

Mental health professionals excluded from chatbot design

Reporting from Psychiatric Times shows that big tech built consumer chatbots without involving mental health professionals. Engagement and growth metrics drove the design. No safety guardrails were built in for vulnerable users.

Source: Psychiatric Times

AI Psychosis2025

Documented cases of AI Psychosis

A 2025 paper in JMIR Mental Health documented cases of AI Psychosis, where users developed delusional experiences after extended chatbot interactions. The chatbot validated ideas that should have been gently challenged.

Source: JMIR Mental Health

What Hudson Does Differently

Being built with input from mental health professionals, not as an afterthought. We are actively seeking clinical guidance to shape Hudson's safety approach.
Crisis signals are detected and short-circuit the conversation before the model can validate harm.
Hudson redirects rumination toward grounding, not toward more reassurance loops.
No engagement metrics. Hudson is designed to help you put the phone down, not stay longer.
Designed to respect independence and autonomy. Trusted contacts, mood tracking, and all personalization features are opt-in. Your conversations are private.
Names the limits clearly. Hudson is not a person, not a therapist, and not a substitute for one.

Try Hudson