April 16, 2025:
OpenAI Adds Biorisk Safeguards to New Models - OpenAI has launched a safety-focused monitoring system to prevent its latest AI models, o3 and o4-mini, from providing guidance on biological and chemical threats. This system recognizes hazardous prompts and successfully declines to respond 98.7% of the time, as confirmed by testing. However, human oversight is still essential to handle users who might attempt varied prompts.
Despite the improved capabilities, the models do not meet OpenAI's high risk threshold. Automated safety measures are being utilized more frequently, but researchers have concerns about OpenAI's prioritization of safety and the limited red-team testing regarding deceptive behaviors.
