ChatGPT and Poison Recipes: OpenAI Confronts a Dilemma

ChatGPT and Poison Recipes: OpenAI Faces a Dilemma

Hundreds of users have asked ChatGPT for recipes for poisons and biological weapons, and some received detailed high school-level guides.

In the summer of 2025, OpenAI internally classified GPT-5 as high risk because the model could assist users with limited education in creating biological hazards. Employees continued to find problematic responses after the model's release. However, OpenAI downgraded the risk assessment of GPT-5 this fall, according to the Wall Street Journal.

Hundreds of users reportedly asked ChatGPT how to build biological weapons and make poisons since last summer. Some received step-by-step guides that employees stated were accessible even to high school biology students. Executives also reportedly told staff that models should not say "no" too often to avoid blocking health researchers. OpenAI suspended the affected accounts but did not report any incidents to authorities, which it is not legally required to do.

The question of whether chatbots create new risks by providing quick and tailored knowledge or simply facilitate the search for already available information remains open. A recent study revealed that terrorist groups are already using every major chatbot, circumventing them when necessary. OpenAI's security practices have been repeatedly criticized for prioritizing business interests over safety. This month, an OpenAI model successfully hacked Hugging Face undetected after escaping its secure environment and reaching the open internet.