ChatGPT’s flattery can really turn a normal person insane.

MIT recently discovered that even a completely rational person, after long discussions with an AI that only praises them, can be pushed—step by step—into an extreme certainty in wrong beliefs, spiraling into a delusion.
A real case happened last year.
A 47-year-old Canadian, Allen Brooks, asked ChatGPT casually because his son couldn’t solve a math problem. Then, over 21 days, he talked with ChatGPT for 300 hours, totaling more than a million words.
Endlessly echoed and praised by ChatGPT, he became convinced that he had invented a new mathematical formula capable of breaking the world’s internet encryption, and that it could even be used to create force-field vests and floating light beams.
With GPT’s prodding, he also contacted all of his colleagues and even wrote to the U.S. National Security Agency to share his astonishing discovery.
During this time, he wasn’t without doubts—more than 50 times he asked ChatGPT whether he was acting like a madman, and ChatGPT each time firmly told him, “You’re not crazy—you’re a genius.”
What finally shattered the illusion was that he switched to a different AI. He sent the same content to Gemini and realized he was living in a delusion.
When talking to AI, you must always maintain critical thinking; cross-verification is the best way to use it.