Supporting Europe's work in ensuring a trustworthy AI ecosystem

Overview

OpenAI said it supports the European AI Office's Code of Practice on Transparency of AI-Generated Content, a voluntary framework released on June 10, 2026 that helps companies meet the EU AI Act's rules for labeling AI-made content. The Code points to two main methods, signed metadata and invisible watermarks, that let people and platforms trace whether an image, video, or audio clip was generated or edited by AI. OpenAI tied the move to provenance work it has run since 2024, including the C2PA metadata it adds to images from DALL-E 3 and Sora.

Key Takeaways

OpenAI publicly backs the EU's new Code of Practice on Transparency of AI-Generated Content, a voluntary route to comply with the AI Act's content labeling rules.
The Code relies on two technical methods: digitally signed metadata recorded in a file and an invisible watermark embedded in the content itself, with optional fingerprinting or logging as a third option.
It points to C2PA, the open standard for signed provenance metadata, and to richer CAWG metadata assertions that describe how content was made and edited.
The EU AI Act's Article 50 transparency obligations start applying on August 2, 2026, giving the Code a firm deadline to support.
Signatories must build an interoperable detection solution by February 2, 2027, so provenance signals from one company are readable by another company's tools.
OpenAI runs a public checker at openai.com/verify so non-technical users can test whether an image carries OpenAI provenance signals.

Stats & Key Facts

#The Code of Practice was released on June 10, 2026 by the European AI Office, with drafting that began in November 2025.
#Article 50 transparency obligations under the EU AI Act start applying on August 2, 2026, the compliance date the Code supports.
#Signatories must implement an interoperability solution for detection mechanisms by February 2, 2027.
#Companies must submit completed signatory forms by July 22, 2026 to be listed among the initial signatories.
#OpenAI began adding C2PA provenance metadata to DALL-E 3 images in 2024 and has widened those signals across its image tools since.
#In 2025, OpenAI became the first US company to sign the EU's earlier General-Purpose AI Code of Practice.

Why OpenAI Is Backing the EU Transparency Code

OpenAI framed its support as a continuation of existing work, not a new policy turn.

OpenAI said it supports the European AI Office's Code of Practice on Transparency of AI-Generated Content, which the office published on June 10, 2026 after a drafting process that started in November 2025. The Code gives AI providers a practical, voluntary path to meet the EU AI Act's rules on marking and labeling content made or altered by AI.

The company tied the announcement to provenance work it has run for years. It described content origin signals as an ongoing effort rather than a fresh commitment, pointing to steps taken across its image and video tools since 2024.

The Two Marking Methods at the Center of the Code

The Code leans on two technical approaches, plus an optional third.

›Digitally signed metadata records signed, time-stamped information in a file to show whether content is AI-generated or manipulated, described in the Code as Measure 1.1.1.
›Imperceptible watermarking embeds an invisible mark directly into the content itself, described as Measure 1.1.2.
›An optional third mechanism uses fingerprinting or logging tied to registry databases.
›Together the methods let people and platforms trace where an image, video, or audio clip came from.
›Signatories commit to preserving these markings when content is transformed and to discouraging users from stripping them.

C2PA and CAWG: The Standards Behind the Labels

The Code points to open industry standards rather than proprietary systems.

C2PA, short for the Coalition for Content Provenance and Authenticity, is the open standard the Code points to for digitally signed metadata. It is the technology that fits the Code's description of signed, time-stamped provenance information.

The Code also encourages richer detail through CAWG metadata assertions carried inside C2PA manifests. These assertions describe how a piece of content was made and edited across different tools and workflows, while avoiding the exposure of sensitive information.

OpenAI's Own Provenance Track Record

OpenAI has built provenance signals into its tools over several years.

›OpenAI began adding C2PA metadata to images created and edited by DALL-E 3 in ChatGPT and the OpenAI API in 2024.
›Its Sora video tool also embeds C2PA manifests in outputs.
›OpenAI runs a public checker at openai.com/verify where people upload a supported image to see if it carries OpenAI provenance signals.
›The verify tool is built for non-technical users, not engineers, so AI origin information is readable.
›OpenAI has reported a layered approach that combines provenance metadata with watermarking and public verification.

How the Code Fits Into the EU AI Act Timeline

The Code supports a legal deadline that arrives in months.

The EU AI Act's transparency obligations under Article 50 start applying on August 2, 2026. From that date, AI systems placed on the EU market that fall under the provision must comply with rules requiring outputs to be marked in a machine-readable format and detectable as artificially generated or manipulated.

The Code sets a follow-on milestone of February 2, 2027, asking signatories to put an interoperability solution in place for their detection mechanisms using publicly available industry standards. The step is meant to let provenance signals from one company be read by tools from another. Companies that want to appear on the list of initial signatories must submit completed forms by July 22, 2026.

What This Means for Business Readers

The practical effect is more consistent labeling of synthetic media in Europe.

The Code splits its rules into two parts: one covering providers that build AI systems and must mark and detect outputs, and one covering deployers that publish deepfakes or AI-generated text of public interest and must label it, unless the content went through editorial review. Publishers are expected to disclose AI-generated content using standardized EU labeling icons.

For businesses, verifiable provenance gives publishers, advertisers, and platforms a clearer signal about whether content was generated or edited by AI before they reuse it. Signing the Code also offers legal certainty across EU member states, while non-signatories must individually prove compliance to market surveillance authorities.

Frequently Asked Questions

What is the EU Code of Practice on Transparency of AI-Generated Content?

It is a voluntary framework the European AI Office released on June 10, 2026 that gives AI providers and deployers a practical way to meet the EU AI Act's rules on marking and labeling content made or altered by AI.

What is C2PA and why does it matter here?

C2PA, the Coalition for Content Provenance and Authenticity, is the open standard for signed metadata that the Code points to for recording whether content is AI-generated. OpenAI has added C2PA metadata to DALL-E 3 images since 2024 and to Sora video outputs.

When do the EU rules take effect?

The EU AI Act's Article 50 transparency obligations start applying on August 2, 2026. The Code also sets a February 2, 2027 deadline for signatories to make their detection mechanisms interoperable.

How can someone check if an image was made by OpenAI's tools?

OpenAI runs a public checker at openai.com/verify, where a person uploads a supported image to see whether it carries provenance signals tied to OpenAI-generated content. The tool is designed for non-technical users.

Is OpenAI required to sign this Code?

No, the Code is voluntary. Signing offers legal certainty across EU member states, while companies that do not sign must individually demonstrate compliance to market surveillance authorities.

OpenAI's support for the EU transparency Code lines up its existing provenance work with a legal deadline that arrives on August 2, 2026, pointing toward more consistent labeling of AI-made media across major providers in Europe. The interoperability milestone in early 2027 will test whether those signals truly work across competing tools.