Full Transparency: Organizations Sign EU Code of Practice on AI Transparency

There was an early show of support for the Code of Practice on Transparency of AI-generated Content (the “Code of Practice”), with around 190 organizations signing up in the first month of its publication. Signing the Code of Practice gives companies a way to demonstrate compliance with their transparency obligations under Article 50 of the EU AI Act (the “AI Act”).

Signatory commitments came not a moment too soon, as most transparency obligations became effective on August 2, 2026, though a grandfathering delay, which means providers that already sold AI systems in the EU before that date have until December 2, 2026 to meet their Article 50(2) marking obligations (under the Digital Omnibus on AI).

What is the Code of Practice?

The Code of Practice works as a voluntary framework for companies to sign up in their role as providers, deployers, or both under the AI Act, and signatories can then rely on the Code of Practice to demonstrate compliance with Articles 50(2), 50(4), and 50(5). Notably, the Code of Practice covers machine-readable marking and detection requirements for providers, perceptible labeling requirements for deployers, and generally requires organizations to align their commitments with best practices. For example, providers must offer free marking-detection tools and ban tampering in their terms or acceptable use policies, while deployers must follow specific label-placement rules depending on the type of content involved. Companies that choose not to sign can still demonstrate compliance in other ways, while using the Code of Practice as a useful benchmark for what regulators expect. The Commission flags in its Q&A that competent authorities (which are different bodies from the Commission itself) “will likely need more detailed information” from non-signatories, who may also face a “larger number of requests for information or access from competent authorities.”

Who has Signed the Code of Practice?

As of September 9, 2026, the signatory tally stands at 82 providers and 152 deployers—drawn from various sectors including tech, telecoms, education and retail—with some companies signing on in both capacities. So far, most signatories are headquartered in the EU and participation has been driven by the private sector, with only a handful of public-sector organizations among the signatories. Roughly half of the signatories are smaller, newer players, but plenty of familiar names have also signed up.

Implementation in Practice

Public statements indicate that some signatories have already implemented, or are in the process of implementing, technical measures to meet the Code of Practice’s provider obligations:

• Google has integrated digital watermarking technology (“SynthID”) into its “generative AI consumer products” and is developing a SynthID Detector “to verify if a piece of content was watermarked with SynthID.” In a July 24 announcement, Google stated its decision to sign the Code of Practice aligns with its “work to adopt and accelerate the C2PA industry standard and to develop AI transparency tools.” Google also noted that it has “been partnering with third-party AI labs … to drive industry-wide adoption of interoperable watermarking tools using SynthID.”
• Anthropic announced on August 14 that “future Claude models will generate text that contains a watermark,” providing “a way of determining the likelihood that Claude was involved in writing the text.” Anthropic stated that it is “implementing this change to comply with the EU AI Act” and will use “a version of the SynthID-Text [watermarking] approach.” Anthropic also expects to “soon be offering a watermark detection API” and in addition to watermarking, Claude supports file metadata generation in accordance with C2PA.
• OpenAI’s current approach to provenance uses “C2PA metadata and SynthID watermarks” for generated images. As announced on July 31, supported audio generated through its tools, including ChatGPT and OpenAI API, “now includes SynthID watermarking,” and OpenAI’s public verification tool “now allow[s] for verification of supported audio files in addition to images.” OpenAI has also stated that it is “working to expand provenance measures for OpenAI’s systems across modalities, including text.”
• Synthesia has stated that videos generated through its platform “carry C2PA-compliant provenance signals that mark them as AI-generated” and that these signals “satisfy the provider-side marking duty (EU Article 50(2)).”

What Other Guidance is Available?

Alongside the Code of Practice, the European Commission has also released its final Guidelines on the transparency obligations under Article 50 (the “Guidelines”). The Guidelines offer practical guidance on how the Commission sees the transparency provisions operating in practice. They explain which AI systems and outputs are in scope and which are exempt (helpfully illustrated with real-world examples), clarify who’s on the hook between providers and deployers, and set out regulatory expectations for compliant technical and user-facing transparency measures.

Looking Ahead

The Code of Practice remains open for organizations to sign at any time, and the first step for those considering signing is determining their role under the AI Act. Organizations can choose to sign the provider or deployer sections (or both) but must accept all commitments within their chosen section(s). Cherry-picking individual commitments is not permitted.

Signatories will be invited to join two task forces expected to launch in September 2026. These task forces will serve as a collaborative forum for participants to exchange industry best practices, share feedback on how the Code of Practice’s transparency measures are working in real time, and help shape the Code of Practice as it continues to evolve.

Maya Vishwanath, an AI Analyst at Morrison Foerster, contributed to this alert.