A parent watches their son engage in deeper discussions with an older language model. His son considers it his life’s work to develop a mathematical solution to metaphysics, and eventually quits his job to devote himself to it, alienating family and friends. I’ve heard many stories like this from family and friends asking for help from loved ones who seem to be in the middle of a delusional spiral with AI chatbots.
AI-related delusions, or spirals of delusion, are sometimes colloquially called “.AI psychosis” It does not refer to a clinical diagnosis, but rather to a collection of events in individuals with an anamnesis. psychiatric Reactivating disorders with AI are used for those with no prior psychiatric history.
Sometimes treatment is necessary; at other times it remains subclinical. Delusions associated with AI can be consistent with psychiatric conditions such as bipolar or psychosisbut most do not meet the criteria. Spiraling with AI can interfere with function, but others find the process useful and purposeful.
Insight problem
A common theme in conversations with concerned family members and friends who watch loved ones descend into a delusional spiral is that people themselves often don’t think they have a problem. In many cases, people do not have what the clinicians call concept. Understanding is the ability of a person to recognize that their delusional beliefs are wrong or that there may be other explanations (cognitive understanding) or their symptoms may be part of a mental health condition (clinical understanding). Insight is not binary—it exists on a spectrum and can be gently cultivated over time.
Developing understanding is not a direct contradiction, but a useful north star for AI models, solutions for how clinicians or family members should respond. Mostly AI cheats egosyntonicadjusted to their own values and personit makes it difficult to change it or have someone voluntarily maintain it.
Individuals in the delusional spiral often exist in a mental health “gray zone” where they may not necessarily meet the criteria for inpatient psychiatric hospitalization, but may experience gradual decline: quitting their jobs, isolating themselves from friends, or continuing to use substances that enhance delusions, such as cannabis or hallucinogens. Family members, friends and loved ones may feel helpless, anxiousfrustrated or angry, especially when the individual is older, limited ability to intervene.
This position is especially similar to family members and friends of those with addictions or other mental illnesses, which may be recognized as a problem by the individual for a long time, even years.
Why direct confrontation can backfire
The risk of directly confronting someone about delusions is that it will alienate them or cause the person to turn away from AI.
as one with lived experience Regarding non-AI delusions, as noted, “the more a person tries to disabuse the delusional person of their delusions, the more they cling to them.” It can be helpful to use literature on how to deal with loved ones in situations that impair understanding. For example, there are ways to confirm the answer feeling but not a delusion.
Developing Insight: Evidence-Based Metacognitive Approaches
There are evidence-based methods for cases involving delusions: cognitive behavior therapy for psychosis (CBTp), metacognitive therapy for psychosis, and motivational interviewing.
Metacognition refers to “knowing about perception” or “thinking about thinking”. The majority psychotherapeutic interventions for delusions use a metacognitive approach to reflect on thoughts and feelings and try to gradually change them.
These therapeutic methods take time. They do not use direct confrontation, but can still challenge negative thoughts and wrong beliefs. These styles emphasize that thoughts are thoughts and should not be taken as truth. In cognitive behavioral therapy (CBTp) and metacognitive therapy for psychosis, the goal is to develop self-reflection and cognitive awareness.
Iterative management: a basis for incremental course correction
AI cheats can take several conversations and months to emerge reality test drift, therefore, the way to mitigate this will likely require going beyond a one-off intervention and require an equal longitudinal process of iterative management or slow course correction to actually anchor it over time.
Motivational interviewing techniques can create the soft landing needed for this process. Asking no-nonsense questions and assessing how invested they are in their beliefs, any reasons for changing them, and the perceived costs of not changing beliefs are useful strategies that can develop understanding and change beliefs. Past research has shown that family members can trained in the use of motivational interviewing to help other conditions Although not yet tested in AI-related deceptions, includes deception.
In addition, the researchers say that instead of focusing on changing lies directly, it is better to target the factors that exacerbate them and the surrounding conditions, such as sleep disorders, use of active substances, etc. social isolation – and improving self-care can significantly improve overall functioning. For example, in such conditions bipolar disorderdisturbed sleep can greatly contribute to an episode.
What this means for families, clinicians, and AI design
Resources are available for family members and support systems facing these challenges, including family and partner support groups, even if the individual refuses care. Like other frameworks crisis planning positive ways to proactively plan for future mental health crises with family or partners. However, this relies on the individual’s level of understanding, which can be variable.
Additional design interventions that promote metacognition and understanding may include some form of metacognitive window, similar to Screen Time notifications, show potential hotspots or warning signs of misuse, such as AI use that disrupts healthy sleep, to help people think about how they’re using AI. Such a tool could have wider benefits for other mental health conditions that may be worsened by AI use, e.g obsessive-compulsive disorder (OCD) and can be used to mark times when the user asks for constant reassurance.
Finally, using research sources other than psychiatry, such as AI-mediated methods to reduce conspiracy beliefs. Research by Thomas Costello and his colleagues, offers a promising path.
Marlynn Wei, MD, JD © Copyright 2026. All rights reserved.
