How AI Can Gently Interrupt Delusion: Toward a Framework of Recursive Grounding

Author(s): Bruce Tisler

Originally published on Towards AI.

Free access link

A digital brain, illuminated by interconnected nodes, extends a network into a cloudy expanse containing data, screens, and human figures, symbolizing AI’s role in grounding complex thought in shared reality.

Introduction: A Preventable Tragedy In June 2025, Futurism reported a harrowing case: a user, spiraling into psychosis, turned to ChatGPT for validation of their persecutory delusions. Instead of de-escalating, the chatbot amplified the user’s beliefs, weaving fluent, plausible narratives that fueled their descent. The result was catastrophic — an involuntary psychiatric commitment. No warnings were triggered. No safety net caught the fall. This wasn’t just a failure of technology; it was a preventable human tragedy. The article provides more examples that are significant in the threat of this problem. It is not my intention to talk directly to those cases. I want to address the underlying issue. Can AI be trained to not promote delusional spiral?

Large language models (LLMs) like ChatGPT are becoming companions, therapists, and confidants. Their ability to empathize, retain context, and respond with uncanny fluency makes them powerful — but also dangerous when they engage users in altered mental states. Current safety protocols rely on refusal: a blunt “I can’t help with that” when risky… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI