GPT-5.5 System Card: SecureBio Found Later Checkpoint 'Performed Highly' Across Biosecurity Evaluations — OpenAI Deploys Strongest Safeguards to Date with Restricted API Rollout

OpenAI·medium signal

OpenAI's GPT-5.5 system card (updated April 24) reveals that SecureBio assessed two pre-release checkpoints from April 2-9, finding the later checkpoint 'performed highly' across biology and biosecurity evaluations compared to leading models. OpenAI withheld API access at launch (April 23) until April 24 citing need for 'different safeguards' — the first time the company has staggered consumer and API access for safety reasons. The system card also documents the first 'High' Preparedness Framework classification for cybersecurity, meaning GPT-5.5 can meaningfully assist with offensive security tasks when misused.