Subscribe
OpenAI
2 min read

OpenAI maps its safety practices to the EU AI Act GPAI Code

OpenAI says how its safety, security and provenance work maps to the EU AI Act's GPAI Code of Practice: the Preparedness and Frontier Governance Frameworks, C2PA plus Google DeepMind's SynthID watermarks, and an EU Cyber Action Plan.

OpenAI maps its safety practices to the EU AI Act GPAI Code

In a July 31 post, OpenAI described how its safety, security, transparency and provenance work lines up with the EU AI Act's General-Purpose AI (GPAI) Code of Practice. The AI Act's transparency obligations under Article 50 apply from August 2, 2026.

OpenAI says it contributed to and endorsed two codes: the GPAI Code of Practice and the Code of Practice on Transparency of AI-Generated Content. The European Commission describes the GPAI Code as a voluntary tool that helps providers of general-purpose AI models comply with the AI Act's obligations on safety, transparency and copyright. The Transparency Code covers the marking and labelling of AI-generated content.

OpenAI points to existing practices that support the GPAI Code: testing models before release, publishing system cards with major releases, bringing outside experts into testing through its Red Teaming Network, and maintaining the public Model Spec, which describes how it shapes model behavior.

Two frameworks sit behind this work. The Preparedness Framework, in place since 2023 and updated in 2025, sets out how OpenAI identifies, evaluates and manages serious risks from advanced AI systems. The Frontier Governance Framework, published on May 28, 2026, explains how those safety and security practices align with legal requirements, including the GPAI Code and California's Transparency in Frontier AI Act. OpenAI says the two frameworks cover risk assessment, safeguards, model reporting, security, incident response, external expert input and ongoing updates.

On provenance, OpenAI uses two signals that it says reinforce each other. Content Credentials, built on the C2PA standard, carry detailed context with a file. SynthID watermarks, a technology developed by Google DeepMind that OpenAI added through a partnership with Google in May 2026, help preserve a signal when metadata does not survive. OpenAI says it is expanding this work from images to audio outputs and working to extend provenance measures across modalities, including text, as standards and tooling mature. It also notes that metadata can be lost, labels can fail to travel across platforms, and no single signal is perfect, which is why it supports a layered approach.

OpenAI uses cybersecurity as its example of governance in practice. Its Trusted Access for Cyber program aims to reduce misuse while helping legitimate defenders use AI. Since launching its EU Cyber Action Plan in early May 2026, OpenAI says it has worked with EU and national cyber agencies, private sector partners and critical infrastructure operators to give them its most advanced cyber models. It says this aligns with the European Commission's Action Plan on Cybersecurity and Artificial Intelligence, which calls for a coordinated approach to the risks of advanced AI while using it to strengthen cyber resilience.

OpenAI also cites the Frontier Model Forum, collaborations with the US Center for AI Standards and Innovation (CAISI) and the UK AI Security Institute (AISI), and work on third-party evaluation practices.

OpenAI says it will keep updating its EU AI Act resources, including model documentation, system cards, usage policies and provenance guidance, as implementation evolves. It lists them in a Help Center article. Teams building on OpenAI models in the EU should treat these documents as a starting point for their own compliance work, not a substitute for it.

Sources

OpenAIPolicyEU AI Act