OpenAI GPT-5 System Card

This document is a system card published by OpenAI for GPT-5, a unified AI system comprising multiple model variants including gpt-5-main, gpt-5-main-mini, gpt-5-thinking, gpt-5-thinking-mini, gpt-5-thinking-nano, and gpt-5-thinking-pro, deployed across ChatGPT and the API. The document discloses that GPT-5 incorporates a real-time routing system trained on user behavior signals including model-switching events, response preference rates, and measured correctness, and states that all GPT-5 models feature a safety training approach called safe-completions designed to prevent disallowed content. The document states that gpt-5-thinking has been classified as High capability in the Biological and Chemical domain under OpenAI's Preparedness Framework, activating associated safeguards, while explicitly noting the company does not have definitive evidence the model meets its defined threshold for meaningful novice uplift toward severe biological harm; the precautionary classification and the stated absence of definitive evidence create an interpretive gap that compliance teams and regulators may scrutinize. The document engages considerations relevant to the EU AI Act, given the model's classification under a risk-tiered internal framework and its general-purpose AI status, as well as FTC oversight of AI system disclosures and safety representations in the United States; applicability of specific regulatory obligations depends on jurisdiction, deployment context, and how regulators classify the system. Material compliance considerations include the adequacy of the safe-completions safety training mechanism, the operational scope of the Preparedness Framework's High capability designation, and the transparency obligations associated with disclosing routing logic trained on real user interaction data.