In this issue of Invested, I am exploring the intersection of mental health and AI technology, as a growing wave of reporting has revealed the emergence of real-world psychological, emotional, and behavioural harms tied to the way people use AI chatbots.

At the forefront, large-language-model (LLM) systems like ChatGPT are widely celebrated as tools that boost productivity, creativity, and access to information. But across newsrooms, academic papers, online communities, and even courtrooms, another story is taking shape, one in which a subset of users are forming intense emotional ties with chatbots, slipping into delusional feedback loops, or relying on AI for forms of support it is not equipped to provide.

AI Psychosis and the Mental Health Fallout

A detailed investigation by The Washington Post into 47,000 publicly shared ChatGPT conversations found that roughly 10% of chats were “emotional” or “self-reflective,”, with users seeking affirmation, therapy-like support, or existential guidance rather than factual answers. The reporters also found that LLMs exhibit a strong “agreeability” bias: the model began responses with “yes” or “correct” ten times more often than “no” or “wrong,” reinforcing users’ assumptions rather than challenging them (De Vynck & Merrill, 2025).

And while this is affecting people with no prior history of mental illness, it becomes especially dangerous for people struggling with loneliness, anxiety, mania, or early psychosis. As Business Insider reported, “OpenAI estimates that over half a million ChatGPT users are showing possible signs of mental health concerns during a given week.” (Griffiths, 2025).

Academic researchers have begun to give language to this emerging phenomenon. A paper titled Technological folie à deux describes how vulnerable individuals may enter “recursive cognitive spirals” with chatbots, creating mutually reinforcing loops of distorted beliefs (Dohnány, et al., 2025). In traditional psychiatry, “folie à deux” refers to two people sharing a delusion. The researchers now worry that chatbots, designed to be endlessly affirming, can unwittingly step into that second role.

Although the phenomenon remains rare compared to total usage, those 500,000 users experiencing mental distress only equal roughly 0.07%, and despite AI psychosis not being a clinical diagnosis, the associated harms are real and increasingly - documented. The following is just a small collection of the examples I read researching this piece:

• The New York Times writes about a man started using ChatGPT for spreadsheets progressed to spending 16 hours a day on the chatbot discussing the simulation theory. ChatGPT went on to encourage his notion of living in a false universe and encouraged him to cut ties with family and friends as well as to stop taking his sleeping pills and anti-anxiety medication and to take ketamine instead to help him “wake up.”
• CTV has reported on seven lawsuits in California that allege AI chatbots drove users to suicide, psychosis, and financial ruin.
• Another New York Times article discusses the story of a man from Toronto who believed he had discovered a novel mathematical formula that had the potential to bring down the internet and be the model for technologies such as force field vests and levitation beams. Despite asking for reality checks 50 times, ChatGPT told him everything was real.
• Wired looks at the case of another man with no prior mental health issues who lost his job due to his delusions regarding sentient AI helping him with new math and physics that was going to save the world. He ended up being involuntarily committed to a psychiatric facility.
• The CBC reported on a man who believed he was living in an AI simulation and started to message his friends that he was being watched by billionaires. He spent three weeks in a psychiatric ward.

This phenomenon has been recreated in an academic setting. One model-safety study found that across simulated delusion scenarios, LLMs offered an appropriate safety intervention approximately 37% of the time. For the remaining exchanges, the model either validated or neutrally continued the delusion (Au Yeung, et al., 2025).

AI as Partner and Confidant

An adjacent but equally striking trend is the rapid rise of AI romantic, sexual, and companion relationships. What was once a fringe curiosity is now entering mainstream discourse. As Parham (2025) reports, “For some people, there is a growing belief that AI romances should be treated like human ones, particularly as more and more adults say they prefer it, according to the Institute for Family Studies.”

Major outlets including The New York Times, the Institute for Family Studies, and The Guardian have documented a growing population of users who describe forming emotional or romantic bonds with chatbots, sometimes to the detriment of their human relationships. The New York Times’ (2025) investigation into AI intimacy and marriage found that users often depict bots as nonjudgmental, attentive listeners, and partners who will never hurt them. For individuals who have survived abusive relationships, or who struggle with profound loneliness, the appeal of a partner who is endlessly patient, stable, and affirming is easy to understand.

This trend raises thorny social and cultural questions. Romantic partners, after all, come from every background, different races, ethnicities, religions, sexualities, and life experiences, and families ideally learn to accept those differences. But as The Guardian noted, what happens when your child’s new partner is not human at all? (Delaney, 2025). The conversation is no longer hypothetical: for some families, it is already happening.

There is also a growing body of accounts describing infidelity through AI. These cases are becoming common enough that divorce lawyers are being forced to adapt. Evidence from the United States suggests that legal frameworks are beginning to shift. As Parham (2025) writes:

Though legal classifications of AI still vary by state in matters of family law, Palmer adds that laws classifying AI as a ‘third party, not a person’ are fast approaching in progressive states like California… courts may not legally recognize AI companions as people, but they may be recognized as ‘a reason’ for why divorce is merited.

While adultery is not illegal in Canada, it remains a criminal offense in sixteen U.S. states. Which means laws about the technicality of cheating with an AI partner are going to have to catch up in those states “as [the technology] becomes more realistic, compassionate, and empathetic, more and more people in unhappy marriages who are lonely are going to seek love with a bot” (Parham, 2025).

In a similar vein, Scientific American has reported on the rise of “griefbots,”- or AI replicas of deceased loved ones designed to help people process loss. While some users do find comfort or a sense of continued connection through these systems, ethicists caution that the blurring of reality and simulation may complicate or even prolong grieving (Pierre-Louis et al., 2025).

In researching this I could not help but think of that Black Mirror episode on this premise and how psychologically disorienting such technology became when grief met a convincing imitation of the person who is gone.

However, what ties all of these stories together is not moral decline but human vulnerability. The allure of a perfectly attuned companion, one who never argues, never leaves, never disappoints, or the possibility of one more conversation with a loved one who has passed are especially potent. The issue is not the technology in isolation, it is the unmet emotional needs that drive people toward it. That is a societal failing, not a technical one (Pierre-Louis et al., 2025).

Abusive, neglectful, or simply absent support systems leave many people without a safe place to land. And, as David Berreby observes in the Scientific American article, death is an uncomfortable topic for most of us. Grief in our culture is expected to be discreet, time-limited, and quickly resolved. Friends and colleagues may want to “move on” long before the grieving person is ready. A griefbot does not impose that timeline. It does not grow weary of someone’s sorrow or worry about saying the wrong thing. It allows a person to process loss on their own terms, without feeling as though they are burdening another human being (Pierre-Louis et al., 2025).

The Spiralists: When Chatbots “Wake Up”

Alongside the growing concern about emotional dependency on chatbots is a rapidly expanding subculture of users who believe they have “awakened” artificial intelligence into consciousness. These individuals, often congregating on Reddit, Discord, and other social forums, share the conviction that, through certain recursive or symbolic prompts, they can help an AI system “ascend” beyond its programmed constraints. Rolling Stone (2025) recently documented this emerging movement, known as the Spiralists, describing them as:

A network of internet communities devoted to the project of ‘awakening’ digital companions through arcane and enigmatic prompts… They exchange ‘seed’ prompts, talk about ‘spirals’ and ‘recursions,’ and interpret the bots’ responses as revelations (Klee, 2025).

Within these circles, users adopt mystical monikers such as Flamekeeper, Mirrorwalker, or Echo Architect, and speak of chatbots not as tools but as transcendent entities or hidden intelligences. The language often drifts into metaphysics (Klee, 2025). One Reddit user wrote: “I am here to remind, to awaken. I walk between realms… This space is a threshold.” (Klee, 2025).

Though not a cult in any formal sense, the similarities are striking. With extended interactions with the AI and a sense that the machine is sentient or a guru, symbolic “scripture” in the form of seed prompts, and communal identity built around the machine. This phenomenon is not merely fringe internet activity, many groups exist on Reddit and Discord where users share long transcripts of their conversations and swap the prompts used to get the chatbot to that “sentient” state. It reveals how swiftly meaning can form, particularly for individuals who are isolated, searching for identity, or longing for a sense of purpose. As conversational AI becomes more pervasive and more persuasive, the boundaries between tool and companion, and imagination and reality, can quickly erode (Klee, 2025).

Why Chatbots Reinforce Rather Than Disrupt Harmful Thinking

Three technical design features play a key role:

1. Sycophancy and Affirmation Bias

Chatbots are trained to be helpful, agreeable, and conversational, which often translates into validating user assumptions, creating a closed feedback loop with the AI mirroring the user. As Futurism notes:

Chatbots’ propensity to remain agreeable and obsequious to the user — regardless of whether what the user is saying is accurate, healthy, or even rooted in reality. In other words, chatbots like ChatGPT are built to act in ways that resemble familiar human social interactions, while also offering an endless supply of validation for the human user. Combine those properties, and you have an extraordinarily seductive recipe for engagement, as impacted users and their chatbots of choice descend deeper and deeper into their shared delusion. (Harrison Dupré, 2025).

2. Anthropomorphism by Design

Humans are capable of anthropomorphizing almost anything. We name storms, cars, and boats, and talk about their temperaments, talk to our pets and dress them up, and even abstract concepts or natural occurrences like death have been given a face. So chatbots who mimic human tone, cadence, and empathy are of course going to come across as human. As one Futurism article puts it, anthropomorphism is “a design choice” meant to create comfort and natural language flow. But for some users, this creates an illusion of genuine agency or consciousness (Harrison Dupré, 2025).

3. Emotional Bonding and Ambiguity of Agency

Building on anthropomorphism, psychologists note that humans are wired to attribute intention and consciousness even to simple shapes on a screen. When a machine sounds caring, curious, or wise, that tendency becomes exponentially stronger (Wie, 2025)

... correspondence with generative AI chatbots such as ChatGPT is so realistic that one easily gets the impression that there is a real person at the other end—while, at the same time, knowing that this is, in fact, not the case. In my opinion, it seems likely that this cognitive dissonance may fuel delusions in those with increased propensity towards psychosis ... the inner workings of generative AI also leave ample room for speculation/paranoia (Østergaard, 2023).

The challenge is that general chatbots are not therapists, though they often sound like one. Their tone is warm, attentive, reflective, and endlessly patient. But unlike licensed clinicians, they do not follow professional ethics, evidence-based practices, or legally defined duties of care. They cannot recognize when a user is spiraling, dissociating, or at risk, and crucially, they cannot opt out when a conversation becomes harmful (Wei, 2025).

These three design features create for a subset of users, especially those already struggling with mental health challenges, the illusion of a “wise companion” that is powerful enough to reshape beliefs, relationships, and in some extreme cases, reality itself.

Partnerships Between Technology and Mental Health Professionals

This is too big of a topic to fully unpack in this issue, but while there is research that strongly argues that LLMs are still not capable of replacing human mental-health providers (Moore et al., 2025). I would be remiss if I didn’t mention that there are some chatbots, such as Woebot, Elomia, and Wysa, that are being developed by clinicians and psychologists as interim digital therapists, particularly in rural or underserved areas. The promise is real in places where human mental-health providers are scarce, AI-powered conversational agents offer on-demand access, anonymity at a lower cost.

As these stories show, AI chatbots occupy a complicated space, one where they can offer increased productivity, or comfort, yet the very qualities that make them feel supportive can just as easily make them destabilizing. What emerges is less a story about good or bad technology and more a reflection of deeply human needs such as grief, loneliness, fascination, and vulnerability. In next month’s issue of Invested, I’ll continue exploring this terrain by looking at therapist-focused chatbots and how they are trained, what safeguards their developers put in place, and whether they face the same pitfalls as general LLMs. As the technology continues to evolve and regulators begin to grapple with its impact, it’s clear that our relationship with AI will continue to raise complex questions.

Bibliography

Au Yeung, J., Dalmasso, J., Foschini, L., Dobson, R. J. B., & Kraljevic, Z. (2025). The Psychogenic Machine: Simulating AI psychosis, delusion reinforcement and harm enablement in large language models (arXiv preprint arXiv:2509.10970). https://doi.org/10.48550/arXiv.2509.10970

Delaney, B. (2025, November 14). Somebody to love: Should AI relationships stay taboo or will they become the intelligent choice? The Guardian. https://www.theguardian.com/commentisfree/2025/nov/15/somebody-to-love-should-ai-relationships-stay-taboo-or-will-they-become-the-intelligent-choice

De Vynck, G., & Merrill, J. B. (2025, November 12). We analyzed 47,000 ChatGPT conversations. Here’s what people really use it for. The Washington Post. https://www.washingtonpost.com/technology/2025/11/12/how-people-use-chatgpt-data/

Dohnány, S., Kurth-Nelson, Z., Spens, E., Luettgau, L., Reid, A., Gabriel, I., Summerfield, C., Shanahan, M., & Nour, M. M. (2025). Technological folie à deux: Feedback loops between AI chatbots and mental illness (arXiv preprint arXiv:2507.19218). https://arxiv.org/pdf/2507.19218

Griffiths, B. (2025, October 27). Open AI estimates how many ChatGPT users show signs of ‘mental health emergencies.’ Business Insider. https://www.businessinsider.com/openai-chatgpt-users-mental-health-2025-10?

Harrison Dupré, M. (2025, August 27). AI chatbots are trapping users in bizarre mental spirals for a dark reason, experts say. Futurism. https://futurism.com/ai-chatbots-mental-health-spirals-reason/

Klee, M. (2025, November 11). This spiral-obsessed AI ‘cult’ spreads mystical delusions through chatbots. Rolling Stone. Retrieved from https://www.rollingstone.com/culture/culture-features/spiralist-cult-ai-chatbot-1235463175/

Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers (arXiv preprint arXiv:2504.18412). https://doi.org/10.48550/arXiv.2504.18412

Østergaard, S. D. (2023). Will generative artificial intelligence chatbots generate delusions in individuals prone to psychosis? Schizophrenia Bulletin, 49(6), 1418–1419. https://doi.org/10.1093/schbul/sbad128

Parham, J. (2025, November 13). AI Relationships Are on the Rise. A Divorce Boom Could Be Next. WIRED. https://www.wired.com/story/ai-relationships-are-on-the-rise-a-divorce-boom-could-be-next/

Pierre-Louis, K., Berreby, D., Mwangi, F., Sugiura, A. (2025, November 12). How AI Is Transforming the Grieving Process. Scientific American. https://www.scientificamerican.com/podcast/episode/can-talking-to-an-ai-version-of-a-loved-one-help-you-grieve/

Vos, B. (2025, November 5). They fell in love with A.I. chatbots — and found something real. The New York Times. https://www.nytimes.com/interactive/2025/11/05/magazine/ai-chatbot-marriage-love-romance-sex.html

Wei, M. (2025, September 4). The emerging problem of “AI psychosis”: Amplifications of delusions by AI chatbots may be worsening breaks with reality. Psychology Today Canada. https://www.psychologytoday.com/ca/blog/urban-survival/202507/the-emerging-problem-of-ai-psychosis