Advancing Responsible AI Across Europe

OpenAI announced that it has strengthened its safety, security, transparency, and provenance framework for developing and deploying responsible AI in Europe, in line with the next phase of the EU AI Act.

OpenAI has endorsed and signed the EU's General-Purpose AI (GPAI) Code of Practice and the AI-generated Content Transparency Code of Practice. Both guidelines, developed through multi-stakeholder discussions, cover safety, security, accountability, transparency, and content provenance management.

OpenAI has been examining model behavior and risks through pre-release testing and public system cards, a Red Teaming Network involving external experts, and the public Model Spec. The Preparedness Framework, introduced in 2023 and revised in 2025, along with the Frontier Governance Framework, specify risk assessment, protective measures, model reporting, security, incident response, and external expert involvement.

OpenAI is also participating in establishing industry-wide safety research and evaluation standards through the Frontier Model Forum, collaboration with the US and UK AI Safety Institutes, and third-party evaluation research.

For provenance management of AI-generated media, OpenAI uses two technologies in parallel.

• Content Credentials (C2PA): Records detailed context in content, such as creation and editing history.
• SynthID watermarking: Helps preserve signals of AI-generated content even when metadata is removed.

OpenAI is expanding the scope of provenance management from images to audio outputs, and plans to continue expanding related measures in line with the EU's transparency guidelines.