OpenAI GPT-5.5 System Card

This document describes the safety testing OpenAI performed on GPT-5.5 and GPT-5.5 Pro before making them available, including adversarial testing focused on cybersecurity and biology risks. OpenAI states these models carry its strongest safeguards to date. GPT-5.5 Pro generally relies on GPT-5.5's safety results rather than always being independently tested.

This system card establishes the predeployment safety process OpenAI applied to GPT-5.5 and GPT-5.5 Pro before release, including completion of OpenAI's full predeployment safety evaluation suite and its Preparedness Framework, with targeted red-teaming focused on advanced cybersecurity and biology capabilities. OpenAI deployed the model with what it characterizes as its strongest safeguards to date, designed to reduce misuse while preserving legitimate uses. For GPT-5.5 Pro, OpenAI generally treats GPT-5.5 safety results as strong proxies, conducting separate evaluations only when OpenAI judges that the Pro setting could materially impact relevant risks or the appropriate safeguards posture. Evaluation results described in the card were conducted in an offline setting except where explicitly noted, and the card is a living document that was updated after original publication to add API-specific safeguard information.

For an individual user, this document establishes that GPT-5.5 and GPT-5.5 Pro were released only after passing a defined set of safety evaluations and adversarial red-teaming, particularly in high-risk areas like cybersecurity and biology. The safeguards in place are described as OpenAI's strongest to date, aimed at reducing misuse while keeping beneficial uses accessible. Users interacting with GPT-5.5 Pro should be aware that its safety posture is generally assessed by proxy from GPT-5.5 results, with independent evaluation applied only where OpenAI judges the Pro setting could materially affect risk.