AI Labels Become an Operational Legal Task: OpenAI and the EU Transparency Codes

OpenAI described its participation in the EU GPAI and AI-content transparency codes, plus provenance work combining C2PA credentials and SynthID signals.

Two codes and an operational shift

On July 31, OpenAI described its support for and participation in the EU General-Purpose AI Code of Practice and the Code of Practice on Transparency of AI-generated Content. The European Commission's official GPAI page lists OpenAI among the signatories.

The voluntary GPAI code covers transparency, copyright, and safety and security. The separate content-transparency code addresses machine-readable marking and detection by providers, plus labeling of deepfakes and certain public-interest text by deployers. The Commission says Article 50 transparency obligations apply from August 2, 2026; signing the code is voluntary, but the underlying legal obligations are not.

OpenAI's stated technical approach

OpenAI says it combines C2PA Content Credentials for image context with SynthID watermark signals that may survive when metadata does not, and is extending provenance to audio and eventually other modalities. It also acknowledges that metadata can disappear, labels may not travel between platforms, and no single signal is perfect.

What organizations should prepare

EU-facing services should separately test machine-readable marks at generation, signal survival after editing or transcoding, user-facing labels, deepfake and public-interest disclosure, and audit records. Responsibility must extend beyond the model provider to publication and distribution.

OpenAI's safety practices are company-reported, and signing a code does not automatically prove legal compliance. Scope, exceptions and human editorial responsibility require service-specific legal review.

Official sources