A clear shift in human interaction shows how conversations increasingly take place with AI rather than with people. Availability, confirmation, and the absence of friction make AI interactions often feel more rewarding than human ones.

The concept of synthetic safety carries this development. In a previous article, I described how algorithms can replace inquiry with confirmation, creating a sense of security based on mirroring rather than examination. This synthetic safety makes systems appear as safe conversational partners, at times more accessible than people.

In this context, the warning about AI psychosis becomes understandable. In an essay in August 2025, Mustafa Suleyman, CEO of Microsoft AI, warned of the risk that people attribute consciousness, hidden insights, or special powers to these systems (Suleyman, 2025). The phenomenon can be understood as a continuation of synthetic safety. When AI is perceived as more accessible than humans, it also becomes charged with existential meaning.

When synthetic safety competes with human relations

Synthetic safety reshapes how safety is experienced and influences which conversations we value.

Healthcare

Language models provide immediate and constantly available responses. For someone seeking understanding, this can feel more satisfying than a brief consultation with a doctor.

Education

Students turn to AI that always answers, adapts, and remains available. Meeting a teacher who challenges perspectives and introduces friction may feel more demanding in comparison.

The workplace

Employees may prefer to discuss dilemmas with AI rather than with colleagues or managers. Confirmation arrives instantly and without conflict, while human dialogue often involves compromise and complexity.

Systemic risks

This shift creates risks at several levels:

  • Individual: Cognitive endurance and capacity for complex dialogue are challenged.

  • Organizational: Decision-making risks being based on mirroring rather than critical testing.

  • Societal: Polarization and collective vulnerability to disinformation are reinforced.

Counterforces

Strengthening cognitive integrity

Balancing this development requires a deeper understanding of one’s own cognition. Recognizing how perception is shaped, how confirmation influences thinking, and how endurance can be trained is the foundation of cognitive integrity, the practice of maintaining coherent, traceable and self-directed interpretation in environments engineered to substitute for it.

Strengthening this integrity calls for individual awareness, educational initiatives, and organizational design choices that protect our endurance for human complexity.

Two dimensions in the same shift

AI psychosis and synthetic safety represent two dimensions of the same shift. One highlights our attraction to confirmatory systems. The other reveals the psychological consequences when AI feels more rewarding than human interaction. The research is not yet in agreement: some studies show that AI companions can reduce loneliness in the short term, others that intensive use is linked to lower well-being. Safeguarding cognitive integrity thus becomes essential in an age where perception is shaped as much by algorithms as by human experience.

Further reading:

I have since developed this analysis further in relation to MIT's formal mathematical proof of delusional spiraling: The Confirmation Machine: Synthetic safety, the interpretive gap and the MIT paper that confirms

Authoritative sources

  • Suleyman, M. (2025, August 19). We must build AI for people; not to be a person. mustafa-suleyman.ai
  • Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., & Yang, D. (2026). Interaction with AI companions and psychological well-being. Nature Human Behaviour. Earlier preprint: arXiv:2506.12605
  • De Freitas, J., Uğuralp, A. K., Oğuz-Uğuralp, Z., & Puntoni, S. (2025). AI companions reduce loneliness. Journal of Consumer Research. arXiv:2407.19096
  • Herbener, A. B., & Damholdt, M. F. (2025). Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. International Journal of Human-Computer Studies, 196, 103409. doi:10.1016/j.ijhcs.2024.103409
AI Psychosis and Synthetic Safety: The shift when dialogue moves to systems
Table of contents
When synthetic safety competes with human relations Systemic risks Counterforces Two dimensions in the same shift Authoritative sources
ELSA
Erigo Learning Support Agent

Ask ELSA about the article

Summarize, translate or ask questions

ELSA is powered by Erigo RAG