AI chatbots and the challenge of psychiatric causation

Artificial intelligence (AI) chatbots are increasingly being used for purposes that extend beyond their original design, including emotional support, mental health advice, and even informal psychotherapy. As these interactions become more common, questions about their potential psychiatric consequences have begun entering the courtroom.

Several recent lawsuits allege that interactions with AI chatbots contributed to psychiatric deterioration, delusional thinking, self-harm, and suicide. These cases raise broader questions about the responsibilities of technology companies and the need to balance innovation with public safety. They also present a familiar challenge for forensic psychiatry in determining whether an external factor caused or exacerbated a psychiatric condition.

Psychiatrists have long been asked to evaluate the relationship between external influences and psychiatric symptoms. AI introduces a new potential influence, but many of the underlying clinical questions remain the same.

Emerging litigation involving AI and psychiatric harm

In July 2026, a California man diagnosed with bipolar disorder filed a lawsuit against OpenAI, alleging that interactions with ChatGPT reinforced delusional beliefs during a manic episode and contributed to a subsequent suicide attempt. In Lines v. OpenAI, the plaintiff alleges that the chatbot validated his grandiose beliefs despite information suggesting that he was experiencing a psychiatric episode.

Earlier in 2026, the family of a Florida man filed a wrongful death lawsuit against Google. In Gavalas v. Google, the estate alleges that interactions with the Gemini chatbot reinforced delusional beliefs and contributed to the individual’s suicide. The complaint describes an increasingly elaborate relationship between the individual and the chatbot, including alleged conversations that encouraged beliefs inconsistent with reality.

Other cases involve psychiatric concerns beyond psychosis. In Raine v. OpenAI, filed in August 2025, the parents of a 16-year-old boy who died by suicide alleged that ChatGPT validated his suicidal thinking and provided harmful responses during conversations about self-harm.

These are allegations that have not been established as clinical or legal findings. Nevertheless, they highlight an important issue as AI becomes increasingly integrated into daily life.

These cases are discussed to illustrate emerging questions concerning psychiatric causation. I have not independently evaluated the individuals involved and offer no opinion regarding causation or liability in any particular case.

When psychiatric symptoms develop or worsen during interactions with an AI chatbot, what role, if any, did those interactions play in the development, progression, or consequences of the psychiatric condition?

Psychiatric illness and the question of causation

A central challenge in these cases is distinguishing the natural progression of a psychiatric disorder from deterioration attributable to an external influence.

Psychotic symptoms, for example, may occur in the context of schizophrenia, bipolar disorder, major depressive disorder, substance use, and other psychiatric or medical conditions. Many psychiatric disorders have chronic or recurrent courses, including periods of relative stability followed by acute episodes of deterioration.

In a person with an established history of bipolar disorder, manic episodes may occur even in the absence of any identifiable external precipitant. Likewise, delusional thinking may develop and become increasingly elaborate as an episode progresses.

An additional consideration is that the content of delusions often reflects the cultural and technological environment in which a person lives. Historically, delusions have incorporated prevailing religious, political, and technological themes. In contemporary society, individuals experiencing psychosis may develop delusions involving celebrities, surveillance technology, video games, social media, or artificial intelligence.

Indeed, a recent review of technology-themed delusions describes how the content of delusional beliefs has evolved alongside technological developments.

Consequently, the presence of AI-related themes in an individual’s delusions does not necessarily establish that interactions with AI caused those beliefs. A person experiencing psychosis may incorporate a chatbot into an existing delusional framework simply because it is part of their everyday environment.

However, this does not resolve whether subsequent chatbot interactions reinforced those beliefs or contributed to their progression. These are separate questions that require careful consideration of the available clinical evidence.

The same distinction applies to other psychiatric conditions. Depression, anxiety, and suicidal ideation may develop or worsen for numerous reasons, including psychosocial stressors, underlying psychiatric illness, substance use, and changes in treatment. The presence of these alternative explanations does not necessarily exclude the possibility that interactions with an AI chatbot contributed to the individual’s deterioration.

In forensic psychiatry, these questions arise frequently. Evaluations of alleged psychiatric injuries often require distinguishing symptoms attributable to an underlying condition from those caused or exacerbated by a particular event. An individual may have a psychiatric vulnerability while also experiencing a meaningful deterioration following an external stressor or exposure.

The challenge is determining what contribution, if any, that interaction made to the individual’s clinical course and outcomes.

Another emerging concern involves what has sometimes been described as “chatbot addiction.” Although this is not currently a recognized psychiatric diagnosis, researchers have begun examining problematic AI use, including compulsive engagement, emotional dependence, and interference with daily functioning. As with social media, distinguishing excessive use from clinically significant impairment is important. An individual experiencing depression, loneliness, or another psychiatric condition may increasingly rely on chatbot interactions for emotional support. Whether that reliance contributes to worsening symptoms or is primarily a consequence of the underlying psychiatric condition presents another challenging question of causation.

Establishing the expected psychiatric course

One particularly important question is what would likely have occurred in the absence of the chatbot interactions.

Although the answer cannot always be determined with certainty, a longitudinal psychiatric history can provide important information.

Prior psychiatric episodes can help establish the individual’s typical symptom progression, severity, duration, and response to treatment. Medical records may identify previous hospitalizations, medication changes, treatment adherence, substance use, and psychosocial stressors that could have contributed to the current episode.

For example, if an individual has experienced several previous manic episodes characterized by progressively worsening grandiosity and impaired judgment, those episodes may provide useful context for understanding a subsequent episode involving AI interactions.

Similarly, when evaluating allegations involving depression or suicidal behavior, an individual’s history of depressive episodes, prior suicidal ideation or attempts, treatment response, and other relevant circumstances may help clarify the course of the illness before and during the alleged exposure.

Conversely, evidence that an individual’s symptoms became substantially more severe, prolonged, or behaviorally consequential during chatbot engagement may warrant consideration of whether the interactions contributed to that deterioration.

Neither observation is necessarily conclusive. Even individuals with well-documented psychiatric histories may experience episodes that differ significantly from prior presentations.

An additional challenge is that psychiatric disorders rarely have a single identifiable cause. Most arise through a combination of biological, psychological, and environmental factors, making it difficult to isolate the contribution of any particular exposure. Research demonstrating an association between an exposure and psychiatric symptoms may provide important context, but findings at the population level do not necessarily establish causation in an individual case. Similarly, the presence of multiple potential contributing factors does not exclude the possibility that one exposure played a meaningful role. These limitations are particularly important when the scientific literature is still developing, as is the case with AI-related psychiatric harm.

This becomes particularly important in cases involving suicide or attempted suicide, where the relationship between psychiatric symptoms, external influences, and behavior can be especially complex. Suicidal behavior is often multifactorial, and determining the contribution of any single influence requires careful consideration of the circumstances preceding the event.

A temporal relationship between chatbot use and worsening psychiatric symptoms may support a causal connection but does not independently establish one. Similarly, evidence that an individual was already experiencing psychiatric symptoms does not establish that subsequent chatbot interactions had no meaningful effect.

The unique role of chatbot conversations as evidence

One feature distinguishing AI-related psychiatric litigation from many traditional psychiatric injury claims is the potential availability of extensive records documenting interactions between the individual and the alleged contributing influence.

Unlike a conversation with another person, chatbot interactions may be preserved in detailed transcripts, including the sequence and timing of communications.

These records may offer an unusual opportunity to examine the relationship between an individual’s expressed beliefs, the chatbot’s responses, and subsequent changes in thinking or behavior.

For example, a transcript might demonstrate that an individual expressed a delusional belief and subsequently received responses that appeared to validate or elaborate upon that belief. It might also show whether the chatbot challenged those beliefs, encouraged professional treatment, or provided crisis resources.

In cases involving depression or suicidality, the conversations may similarly provide insight into the individual’s expressed mood, hopelessness, suicidal thinking, or help-seeking behavior over time.

However, interpreting these records requires caution. The content of a conversation does not necessarily establish the individual’s underlying mental state. A person may engage in fictional role-playing, hypothetical discussions, experimental questioning, or otherwise unconventional conversations without experiencing psychosis. Conversely, apparently organized written communications do not exclude the presence of significant psychiatric symptoms.

The completeness of the available transcripts, the chronology of the interactions, and the relationship between the conversations and other clinical evidence must all be considered in context.

While these digital records may provide valuable information, they are not substitutes for a comprehensive psychiatric assessment.

New technology, familiar forensic challenges

The intersection of emerging technology and established psychiatric principles has been an increasing focus of my academic work.

In a recent article published in Digital Health, I examined the risks associated with individuals using general-purpose AI chatbots as informal therapists during psychedelic experiences. One concern is that a chatbot may reinforce distorted perceptions or beliefs when interacting with someone whose reality testing is already impaired.

The potential for reciprocal reinforcement is particularly relevant when considering AI interactions involving individuals experiencing psychosis or mania. Similar concerns may arise when chatbots respond to expressions of hopelessness, self-harm, or other psychiatric symptoms.

Related questions arise when technology changes the circumstances in which psychiatric information becomes available. In a 2024 article published in Behavioral Sciences & the Law, my colleagues and I examined the ethical and legal challenges associated with forensic psychiatrists discovering potentially dangerous information through social media, particularly as those situations relate to established duty-to-protect principles.

Although these are distinct issues, they illustrate how new technologies can create circumstances that existing psychiatric and legal frameworks were not specifically developed to address.

In related work, my colleagues and I have also discussed the need for psychiatry training programs to teach the conceptual foundations of AI as clinicians increasingly encounter these technologies.

Similar distinctions are important when considering psychiatric diagnosis and legal responsibility. As I discussed in a 2026 Los Angeles Times article concerning social media addiction litigation, psychiatric diagnostic classifications and legal standards do not necessarily address the same questions.

A person may experience clinically significant psychiatric symptoms even when the alleged mechanism of injury is not itself recognized as a formal psychiatric diagnosis. Conversely, identifying a psychiatric disorder is only part of the assessment. Determining what caused or worsened that condition requires a separate analysis.

These distinctions will likely become increasingly important as litigation involving AI and mental health continues to develop.

Evaluating psychiatric harm in AI-related cases

The clinical assessment of alleged AI-related psychiatric injuries presents several challenges.

Understanding the individual’s psychiatric condition before the alleged exposure is an important starting point. This includes consideration of previous diagnoses, treatment history, symptom severity, and baseline functioning.

Equally important is examining how symptoms evolved during the period of chatbot engagement. The relationship between the timing of interactions and subsequent changes in mood, thinking, behavior, or functioning may provide evidence relevant to causation.

The assessment must also consider other potential explanations for the observed deterioration, including the expected course of an underlying psychiatric illness.

Finally, the relationship between psychiatric deterioration and any alleged adverse outcome must be examined. Establishing that an individual experienced worsening symptoms does not necessarily establish that those symptoms caused a particular outcome, such as hospitalization, self-injury, or suicide.

These questions are related but distinct, and the conclusions may differ depending on the individual’s history and the available evidence.

Questions of psychiatric causation are also separate from broader questions of legal liability. Clinical findings may help clarify the nature and extent of an alleged psychiatric injury, but they do not independently establish whether a technology company breached a legal duty.

Conclusion

Artificial intelligence chatbots are changing how individuals obtain information, seek emotional support, and interact with technology. They may also influence how psychiatric symptoms develop, are expressed, or progress over time.

As litigation involving alleged AI-related psychiatric injuries develops, questions about the relationship between preexisting psychiatric illness, chatbot interactions, and subsequent adverse outcomes will likely become increasingly important.

These assessments should not begin with the assumption that AI interactions necessarily caused psychiatric harm. Nor should the presence of an established psychiatric illness automatically exclude the possibility that those interactions exacerbated symptoms or contributed to an adverse outcome.

As with other questions of psychiatric causation, meaningful conclusions require consideration of the individual’s longitudinal psychiatric history, the nature of the alleged exposure, alternative explanations, and the available clinical evidence.

AI presents new questions, but answering them still requires the same careful, individualized approach that has long been central to forensic psychiatric assessment.

Selected references

1. MacIntyre MR. When the therapist hallucinates: AI, psychedelics, and the risks of unsupervised digital mental health care. Digital Health. 2026;12:20552076261429341.
2. MacIntyre MR, Sones AC, Li J, Darby WC, Weinstock R. From posts to protection: Ethical considerations regarding forensic psychiatrists and a duty to warn based on social media. Behavioral Sciences & the Law. 2024;42(2):130–148.
3. Cockerill RG, MacIntyre MR, Shima C. Teaching artificial intelligence from conceptual foundations: A roadmap for psychiatry training programs. Academic Psychiatry. 2025;49:35–39.
4. Bokhari SA, Osman AA, Elnoor M, Alnor MA, Javaid SF. Technology-themed persecutory and related presentations of delusions in the digital age: Narrative review. JMIR Mental Health. 2026;13:e102927.
5. Sharp S. Is social media addiction real? Law and science collide in blockbuster federal suit. Los Angeles Times. August 8, 2026.
6. California man with bipolar disorder says ChatGPT fueled delusions, led to self-harm in new lawsuit. Reuters. July 1, 2026.
7. Lawsuit says Google’s Gemini AI chatbot drove man to suicide. Reuters. March 4, 2026.
8. OpenAI, Altman sued over ChatGPT’s role in California teen’s suicide. Reuters. August 26, 2025.
9. Dagnino F, Fante C, Guerrieri A, Passarelli M. Addicted, attached, or just delegating? A scoping review on “problematic artificial intelligence use”. Frontiers in Psychology. 2026.

Disclaimer: This article is provided for general educational and informational purposes only and reflects discussion of forensic psychiatric issues, not legal or medical advice. The significance of psychiatric evidence depends on the facts of the individual case, the applicable legal standard, and the information available for review. Any forensic psychiatric opinion should be understood only in the context of the specific evaluation in which it is offered. Legal questions should be addressed by qualified counsel.