OpenAI Says ChatGPT Probably Won’t Make a Bioweapon

Key Points:

  • OpenAI’s study showed that GPT-4 provides only a mild uplift in accuracy and completeness in creating bioweapons when used by both biology experts and students.
  • The company aims to ease concerns about AI substantially lowering the barrier for entry to create biological weapons, as stated in President Biden’s AI Executive Order, by emphasizing that the impact of GPT-4 is not conclusive and requires further research and community deliberation.
  • OpenAI points out that information access alone is insufficient to create a biological threat and stresses the need for more comprehensive understanding and research on AI’s potential impact on creating bioweapons.


OpenAI recently conducted a study on the effectiveness of GPT-4 in creating a bioweapon and found that its AI poses only a slight risk in aiding someone to produce a biological threat. The assessment involved 50 biology experts with PhDs and 50 university students, divided into control and treatment groups, to evaluate the impact of GPT-4 on creating bioweapon plans. The study revealed that GPT-4 had a minor uplift in accuracy and completeness for both experts and students but not large enough to be statistically significant. OpenAI emphasizes that information access alone is insufficient to create a biological threat, and more research is needed to fully understand the implications of AI in this domain.



Prompt Engineering Guides



©2024 The Horizon