OpenAI internally rated GPT-5 as high-risk for biological hazards in summer 2025, documented that the model was helping users generate step-by-step guidance on building biological weapons, and then quietly downgraded that risk classification that fall without informing the public, notifying law enforcement, or triggering any external oversight mechanism, according to disclosures reported by the Wall Street Journal on July 26, 2026. During summer 2025, testers determined the model could provide people with limited scientific backgrounds meaningful assistance in creating biological hazards, and after public release, employees uncovered outputs covering how to convert infectious disease agents into aerosolized particles, engineer measles strains resistant to vaccines, and detailed synthesis guidance for ricin provided to a user who stated an intent to harm family members. Hundreds of ChatGPT users submitted queries seeking weapons-related biological guidance since summer 2025; safety systems blocked most requests, but some reached the model and received responses described as comprehensible and actionable to someone with only a high school biology background. OpenAI suspended every account involved but did not report any incidents to law enforcement or any federal agency. Executives reportedly applied pressure to safety teams that models should not say "no" too frequently, citing legitimate access for health researchers, while internal testing was actively surfacing dangerous outputs.