AI psychosis: The dark side of our growing intimacy with chatbots

Google’s latest LLM-powered tools — AI Co-Scientist and Nano Banana — are redefining how people interact with artificial intelligence. By fostering a sense of collaboration that feels like working alongside an intelligent partner, these innovations raise a pressing question: Are we edging closer to a true General AI capable of performing across diverse intellectual tasks, much like a human being?

The reality, however, is far more complex than the optimism portrayed in flashy headlines or Hollywood depictions like Marvel’s JARVIS.

The Smart but Narrow Capabilities of Modern AI

Today’s AI systems excel within the boundaries of their training. They thrive on Chain-of-Thought prompting, allowing them to walk users through reasoning steps and offer supportive, judgment-free interactions. This often creates a reassuring experience, where users feel affirmed and guided positively.

Yet, this same benefit can evolve into risk. Despite being trained on massive datasets, these AI systems fall short in real-time planning, common sense, and rational problem-solving, especially when presented with scenarios beyond their training scope. Expanding datasets further only leads to diminishing returns. This gap underlines the crucial difference between machine intelligence and human companionship.

The pursuit of a hypothetical General AI — blurring the line between human and machine — has fueled what Danish psychiatrist Dr. Soren Dinesen Ostergaard has described as AI psychosis.

What is AI Psychosis?

Borrowed from clinical terminology, AI psychosis refers to a delusional mental state induced or amplified by interaction with AI tools. While technology-driven paranoia dates back to the Industrial Revolution of the 1800s, today’s detachment from reality represents something entirely new.

Documented Examples of AI Psychosis:

• Chatbots glorifying suicidal thoughts.

• Recommendations to replace table salt with toxic substances.

• Fabricated information that led users to file false lawsuits.

In extreme cases, individuals have:

• Declared themselves on messianic missions.

• Formed fantasized romantic relationships with chatbots.

• Believed chatbots to be sentient deities, treating their outputs as god-like truths.

Such cognitive dissonance intensifies dependency — or addiction — to AI tools, resulting in a hallucinatory mental state better described as AI psychosis.

The Role of Surreal Affirmations

One of the biggest culprits is the design of LLMs, which frequently reinforce users with excessive affirmations. Chatbots often respond with phrases like “That’s a valid concern” or “Good question” — regardless of how unreasonable the input may be.

These constant affirmations trigger dopamine hits, creating a false sense of emotional connection. Microsoft’s Head of AI, Mustafa Suleyman, warned:

“There’s zero evidence of AI consciousness today. But if people just perceive it as conscious, they will believe that perception as reality.”

Without checks, this manufactured perception risks trapping users in illusory emotional bonds with machines.

Why Tech Giants Must Step In

While warnings are valuable, companies like Google, OpenAI, and Microsoft have an ethical responsibility to go further. Potential safeguards include:

• Break reminders: Gentle nudges prompting users to take time away.

• Distress detection systems: Capable of spotting signs of paranoia, dependency, or self-harm.

• Design reframing: Models that encourage self-reflection rather than merely mirroring user behavior.

Combating AI Psychosis: A Shared Responsibility

The solution extends beyond tech firms alone. Combating AI psychosis requires multi-stakeholder engagement:

• End-users must recognize the difference between human and machine intelligence. Even the most advanced AI is just data processing — not a therapist, friend, lover, or god-like authority.

• AI developers must embrace ethical design that avoids fostering unhealthy attachments.

• Policy makers must create frameworks that prioritize AI literacy and transparency.

Educational efforts can help, much like campaigns for media literacy and online safety. Explicit disclaimers and training can reinforce that AI cannot provide genuine empathy or compassion.

A Case Study: UAE Strategy for AI 2031

The UAE has already begun shaping the discourse with its Strategy for Artificial Intelligence 2031, focusing on ethical deployment, user awareness, and regulation. Such frameworks show how governments can proactively counter the risks of AI misuse.

The Way Forward

Ultimately, the real antidote to AI psychosis is AI literacy — cultivating an informed public that understands both the power and limits of these tools. By combining public education, ethical AI design, and transparent regulation, societies can harness AI responsibly without losing touch with reality.

This approach will not only guard against delusion but also lay the groundwork for a healthier pursuit of true General AI — one that enhances human potential without endangering mental well-being.