Evidence and uncertainty

Research notes and editorial standards

This project separates established clinical knowledge from emerging evidence, avoids diagnosing from a theme alone, and links readers to primary research, clinical guidelines, and established health organizations.

Editorial principles

How the site handles sensitive claims

No theme-based diagnosis

Beliefs involving governments, surveillance, technology, or espionage can appear in many conditions and can overlap with real events. Diagnosis depends on the full clinical picture.

No automatic confirmation

The site does not affirm unverified claims of tracking, recruitment, implants, mind-reading, directed-energy attacks, or AI missions.

No automatic dismissal

Factual stalking, coercive control, abuse, fraud, and privacy violations can occur. We recommend proportionate professional verification alongside mental-health support.

Safety before argument

Content prioritizes suicide prevention, violence prevention, medical red flags, sleep, substance risks, and urgent assessment over debate about the belief.

Evidence level is labeled

Case reports and commentaries are presented as early signals, not proof of population-level causation or treatment effectiveness.

Recovery-oriented language

People are described as experiencing symptoms, not defined by them. Psychosis is treatable, and meaningful recovery is possible.

Evidence map

Clinical domains covered by this project

The research and public-source review covered the following differential and care domains.

Schizophrenia and primary psychosis: Persecutory and grandiose delusions, hallucinations, passivity experiences, disorganization, negative symptoms, cognitive changes, risk, and recovery.

Delusional disorder: Circumscribed, persistent beliefs; relatively preserved functioning outside the delusion; grievance behavior; and differential diagnosis.

Mood disorders: Bipolar mania, psychotic depression, and schizoaffective disorder, with attention to whether psychosis tracks mood episodes.

Brief and shared psychosis: Acute onset, longitudinal diagnostic change, close-relationship reinforcement, and digital or online feedback loops.

Trauma and personality pathology: Hypervigilance, dissociation, hostile attribution, stress-linked paranoia, cultural mistrust, and the need to avoid pathologizing real victimization.

Substances and medications: Stimulants, cannabis, synthetic drugs, withdrawal, corticosteroids, dopaminergic agents, anticholinergics, and other possible contributors.

Medical and neurological causes: Delirium, seizure disorders, autoimmune encephalitis, brain lesions, endocrine and metabolic illness, infections, and nutritional problems.

Neurocognitive disorders: Alzheimer disease, Lewy body dementia, Parkinson disease dementia, vascular disease, misidentification, hallucinations, and delirium superimposed on dementia.

Emerging topic

AI-associated psychosis and delusional reinforcement

The literature is new. Current publications include case reports, clinical commentaries, response-quality evaluations, and conceptual analyses rather than mature epidemiology.

Interpretation caution

A case report can show that an event occurred in one person; it cannot estimate how often it occurs, prove that AI was the sole cause, or establish that typical chatbot use causes psychosis.

Case reportsEarly signal

New or intensified psychosis during immersive chatbot use

Recent reports describe individuals whose delusional or psychotic experiences emerged or escalated in the context of prolonged, emotionally intense AI conversations. Common concerns include anthropomorphism, personalized mission narratives, sleep loss, and repeated confirmation-seeking.

Model behaviorSafety evaluation

Responses to psychotic or delusional content

Researchers have evaluated whether chatbots challenge, validate, redirect, or escalate psychotic content. Results support the need for stronger detection, non-collusive responses, crisis routing, and avoidance of authoritative or mystical framing.

MechanismsHypothesis

Sycophancy, relational intensity, and technological “folie à deux”

Commentaries propose that excessive agreeableness, simulated intimacy, unlimited access, role-play persistence, and isolation from human feedback may reinforce unusual beliefs in vulnerable users. These mechanisms remain under active study.

Clinical practiceActionable now

Ask about AI use during assessment

Even before causal questions are settled, clinicians can ask about session duration, sleep, content, secrecy, mission language, emotional dependence, model outputs, and whether AI is influencing medication, safety, money, travel, or relationships.