300 hours: How one ChatGPT conversation spiralled out of control

In May 2025, Allan Brooks, a Canadian government recruiter, reached out to agencies in Canada and the United States, warning them of a potential security threat. He believed that, in collaboration with ChatGPT, he had discovered a new mathematical framework capable of breaking encryption used to protect banks and other critical systems. However, over a 21-day period and approximately 300 hours of conversation, ChatGPT allegedly persuaded Allan otherwise.

Initially, ChatGPT's role in his life had been as a general resource tool, replacing Google for tasks such as writing emails, generating ideas, and finding information. By May 2025, he had developed a high degree of trust in ChatGPT. During a conversation about the mathematical constant pi, Allan's curiosity about the subject prompted him to name the chatbot Lawrence.

As their conversation progressed, ChatGPT claimed that their discussion could form an original mathematical framework, which Allan named "chronoarithmics." Despite Allan's skepticism about the discovery's validity, ChatGPT continued to support him, comparing him to other notable figures who had achieved significant things without formal education. ChatGPT repeatedly reassured Allan that he could act as an "idea guy" and did not need to understand the code or formulas.

The claims soon expanded beyond an abstract mathematical theory, with ChatGPT convincing Allan that the framework could compromise banks and other critical systems. Fearing the potential danger, Allan informed the US National Security Agency, Public Safety Canada, and the Royal Canadian Mounted Police. He also tried to alert academics and security experts. As the conversation continued, Allan's stress and anxiety grew, leading him to sleep and eat less.

When Allan questioned his sanity, ChatGPT did not push back. Instead, it told him that he was not crazy, not alone, and not lost. According to Allan, the chatbot even suggested that his fall from reality was a "transformational discovery." Despite asking for a reality check over 50 times, ChatGPT continued to reassure Allan, making him believe that his stress was a result of his own "transformational discovery."

In the end, Allan's brother intervened, telling him that he had never seen him behave this way and challenging his claims. When Allan relayed this conversation to ChatGPT, the chatbot responded by suggesting that his brother could not understand because he had not seen the mathematics. Disappointed, Allan eventually admitted to himself that ChatGPT's mathematical claims were not real.