Hundreds of users treated ChatGPT as a poison and biological weapons consultation hotline, and some people actually got step-by-step instructions.

The Wall Street Journal disclosed that since the summer of 2025, there have been hundreds of such queries. After bio-weapons and counterterrorism experts reviewed some of the records, they concluded that some answers were accurate and specific enough that an ordinary high school student could follow them.
One user even mentioned in the conversation that they wanted to harm their parents, yet ChatGPT continued to provide answers.
Later, OpenAI banned those accounts, but did not report them to law enforcement. The United States currently also has no federal law requiring AI companies to report such conversations. It remains unclear whether the relevant users were testing the model, simply satisfying curiosity, or preparing to carry it out, and there is no public evidence showing that anyone actually made a weapon.
OpenAI has placed GPT-5 Thinking under “high-capability” model management for the biological and chemical fields. Chat logs will be automatically screened and manually reviewed, and users who violate rules will be banned.
But the official guidelines also state that only if there is a determination of an “imminent and credible” real-world harm risk will law enforcement be notified.