OpenAI Frontier Governance Framework

This document is OpenAI's Frontier Governance Framework, a voluntary corporate governance disclosure that establishes OpenAI's stated safety, security, and risk management practices for frontier AI development; it does not function as a user-facing contractual agreement but rather as an institutional policy statement describing internal governance structures and their alignment with emerging regulatory frameworks. The framework states that OpenAI maintains a Preparedness Framework to assess catastrophic risk across four domains (cybersecurity, CBRN, persuasion, and autonomous replication), a Safety Advisory Group, and a tiered model risk classification system (low, medium, high, critical) that the terms describe as governing deployment decisions for frontier models. Notably, the document asserts voluntary alignment with the EU AI Act and California SB 1047-style requirements before those frameworks became legally binding on OpenAI, and it establishes hardcoded behavioral restrictions described as absolute prohibitions that the policy states cannot be overridden by any operator or user instruction, which is operationally distinct from typical terms-of-service acceptable use clauses in scope and framing. The framework explicitly engages the EU AI Act (including provisions on general-purpose AI models with systemic risk), California SB 1047 (as a reference point for safety commitments), and voluntary commitments made to the White House and UK AI Safety Institute. Material compliance considerations include whether the voluntary commitments described constitute enforceable obligations under applicable law, how the stated incident reporting and third-party auditing commitments interact with emerging mandatory disclosure regimes, and how the framework's stated alignment with the EU AI Act will be operationalized as that regulation enters into force; these dependencies are jurisdiction- and timeline-specific.