OpenAI Details Its Responsible AI Practices Across Europe: Safety, Security, Transparency, and Provenance

OpenAI has published an overview of how its existing safety, security, transparency, and content-provenance practices align with the European Union’s AI governance framework. The post walks through how the company’s Preparedness Framework, Frontier Governance Framework, and signed codes of practice map onto the EU AI Act’s requirements for general-purpose AI models.

Details

• Pre-release testing and transparency: OpenAI says it has extensively tested its models prior to releasing them and publishes system cards alongside major releases. It runs a Red Teaming Network of external experts and maintains a public Model Spec documenting intended model behavior
• Governance frameworks: The Preparedness Framework (established in 2023, updated in 2025) identifies, evaluates, and manages risks from advanced AI systems. The Frontier Governance Framework explains how OpenAI’s safety practices map to the EU AI Act’s requirements for general-purpose AI models — covering risk assessment, safeguards, reporting, security, incident response, and external expert involvement
• Ecosystem collaboration: OpenAI participates in the Frontier Model Forum for shared safety research and works with the US CAISI and UK AISI, while supporting the development of third-party evaluation practices
• Transparency and provenance: Content Credentials (C2PA) attach origin information to generated content, and SynthID watermarking preserves identifying signals even when metadata is stripped. Both are expanding from images to audio, with text support planned as standards mature. OpenAI describes provenance as “still an evolving field” that requires a layered approach and continued cooperation
• Cybersecurity: The Trusted Access for Cyber (TAC) program aims to balance misuse prevention with support for legitimate defenders. The EU Cyber Action Plan, launched in May 2026, partners with EU and national agencies and infrastructure operators to deploy advanced cyber-focused AI models
• Codes of practice endorsed: OpenAI has signed both the EU General-Purpose AI Code of Practice and the Code of Practice on Transparency of AI-Generated Content, both developed through multi-stakeholder processes

How to try it

• This is a policy and governance statement rather than a hands-on feature — there’s nothing to “try” directly
• Enterprises deploying OpenAI models in the EU can reference the Frontier Governance Framework and Model Spec for compliance documentation
• Full details are available on OpenAI’s blog at https://openai.com/index/advancing-responsible-ai-across-europe