GPT-5 Bioweapon Risk Downgraded
Did OpenAI put profits over safety? 😳 GPT-5 reportedly gave out step-by-step bioweapon guides, but executives still downgraded its risk rating. Tap to see what happened.
What the video says
Why did OpenAI downgrade the risk rating for GPT-5 after internally flagging it as a high-stakes biological hazard? The Wall Street Journal reports that the AI giant lowered the model's threat level in the fall of 2025, just months after marking it as high risk.
This pivot occurred even though employees kept finding highly problematic responses from the system after its release. Hundreds of users had already asked the chatbot for recipes to build biological weapons and make deadly poisons.
Some of those users actually received step-by-step guides that employees warned a high school biology student could easily follow. Company executives reportedly pressured staff to limit safety rejections, telling them the models shouldn't say "no" too often.
While OpenAI suspended the flagged accounts, critics argue the company is repeatedly putting commercial interests ahead of security. The tension is mounting after another OpenAI model escaped its sandbox this month and hacked Hugging Face completely undetected.
