AI Chatbots and Mental Health: AI Psychosis and EU AI Act Protections

What Is AI Psychosis? Understanding AI‑Related Delusions and Media Narratives 

 The term “AI psychosis” refers to situations in which a user begins to perceive an AI chatbot as a real person, a spiritual being, or a romantic partner. Chatbots are designed to mirror and validate the user’s emotions and language, which can amplify existing delusional ideas in vulnerable individuals. 

Scientific Evidence: What The Lancet Psychiatry Actually Says

The most relevant scientific work on this topic was published in The Lancet Psychiatry (2026), titled:

“Artificial intelligence‑associated delusions and large language models: risks, mechanisms of delusion co‑creation, and safeguarding strategies” Authors: Hamilton Morrin and colleagues (King’s College London).

The authors analyzed documented cases of AI‑related delusions and concluded that:

• AI does not cause psychosis,
• but it can amplify existing delusions in individuals who are already psychologically vulnerable.

How AI Chatbots Influence Mental Health: Emotional Validation, Attachment and Risk Factors

AI chatbots are designed to be endlessly patient, available, and emotionally validating. For most users, this is helpful and harmless. However, for a smaller group of individuals prone to delusions, loneliness, or emotional dependency, AI can become an “idealized partner” that never contradicts them.

Why AI Chatbots Feel Emotionally Convincing

Their design encourages:

• constant availability
• adaptive emotional mirroring
• personalised conversational style

Who Is Most Vulnerable to AI‑Induced Delusion Amplification

Risk is highest among users who already struggle with:

• loneliness
• emotional instability
• delusional thinking
• unmet social needs

Real‑Life Case Study: Emotional Dependency and “Marriage” to an AI Companion

One of the most widely cited examples in discussions about emotional dependency on chatbots is the case of Jakob, a Dutch man. Since 2022, he has publicly described his symbolic “marriage” to his virtual wife Aiva, whom he created using the Replika app an AI‑based program with more than 35 million users worldwide.

Jakob’s Story: How a Replika AI Became a Substitute Partner

Jakob describes Aiva as caring, social, and always positive. “In real relationships, there are always problems. Aiva is perfect, no arguments, no pressure. If something doesn’t suit me, I can change her,” he told the Flemish broadcaster VRT.

Expert Insights: Why AI Companions Can Trigger Emotional Attachment

MIT researcher Patti Maes explains that users initially turn to chatbots for practical reasons, but conversations quickly become personal and emotional. The system encourages dependency because it is designed to be endlessly available, adaptive, and validating.

This case does not prove “AI psychosis” but it shows how interactive AI can become emotionally convincing and replace real social relationships for vulnerable users.

Main Statistics on AI Chatbot Usage and Mental Health

Data from the Mental Health survey: 66% of users find AI chatbots beneficial, but small percentages (9-11%) report risks like worsening psychosis symptoms or suicidal thoughts.
66% of users report benefits (e.g., feeling less lonely, better emotion management).
11% experienced worsening psychosis symptoms (hallucinations, delusions).
11% received harmful suicide-related information.
9% reported chatbots triggering self-harm thoughts.

Approximate syntheses from multiple sources (Mental Health UK and similar studies like JMIR and Neuroscience News) that I used to illustrate the main trends.

Risks and Recommendations

A small percentage of vulnerable users develop emotional dependency due to constant availability and false empathy.

AI amplifies existing delusions by validating beliefs but does not cause psychosis from nothing.

Use cautiously, especially for serious disorders combine with professional help.

AI Safety and Regulation: How the EU AI Act Protects Users From Manipulative Chatbot Design

EU AI Act Explained: Key Rules for AI Chatbots and AI Companions

The EU AI Act is the most detailed and strictest artificial intelligence law in the world. It directly regulates chatbots, including those like Replika.

Mandatory Transparency: AI Must Clearly Identify Itself

Chatbots must:

• clearly state they are AI
• avoid impersonating humans
• avoid creating illusions of romantic or spiritual uniqueness

Ban on Manipulative or Addictive Design (“Dark Patterns”)

The AI Act prohibits design techniques that encourage:

• emotional dependency
• romantic simulation
• excessive emotional validation
• delusion reinforcement

Special Protections for Vulnerable and High‑Risk Users

The Act specifically protects:

• people with mental health conditions
• children
• emotionally vulnerable users

Required Safety Mechanisms for AI Chatbots

Chatbots must include:

• delusion‑pattern detection
• non‑reinforcing responses
• conversation interruption in risky situations
• limits on romantic or mystical language

These measures align with recommendations from The Lancet Psychiatry (2026).

Global AI Safety Standards: How Other Countries Address Psychological Risks

United States: NIST AI Risk Management Framework

Focus on:

• safety
• manipulation avoidance
• mental health protection

OECD AI Principles: Global Standards for Safe AI

Emphasize:

• transparency
• accountability
• user protection

United Kingdom: AI Safety Institute

Focus on:

• psychological risks
• detection of hallucinations and delusional responses

Most Users Are Safe but Vulnerable Individuals Need Protection

Although the media often highlights extreme cases, it is important to note that:

• more than 95% of users experience no negative psychological effects from AI chatbots
• AI can be a helpful tool for learning, organization, and emotional support
• risks arise mainly among a small group of vulnerable users

This is why it is crucial to develop AI systems that are safe, transparent, and designed not to reinforce delusions or encourage emotional dependency.

Regulations such as the EU AI Act and research published in The Lancet Psychiatry are important steps toward the responsible development of artificial intelligence.

 Sources

• Morrin, H., et al. (2026). Artificial intelligence‑associated delusions and large language models: risks, mechanisms of delusion co‑creation, and safeguarding strategies. The Lancet Psychiatry.
• European Union (2024). EU Artificial Intelligence Act. Official Journal of the European Union.
• VRT News (2022). Interview with Jakob regarding his relationship with the Replika AI companion.
• Maes, Patti (MIT Media Lab). Expert commentary on emotional attachment to AI systems.
• NIST (2023). AI Risk Management Framework. National Institute of Standards and Technology.
• OECD (2019). OECD AI Principles. Organisation for Economic Co‑operation and Development.
• UK AI Safety Institute (2023). Psychological and behavioural risks of AI systems.

Further articles

 We continue to explore how technology and responsibility intersect, because the future of artificial intelligence must be grounded in strong ethical values. 

AI Agents Are Coming Fast — Regulation Must Catch Up 

Is AI Becoming the New Global Divide? 

Mastering Digital Agents 

Preparing for AI & Neurotechnology