Meta signs EU AI Act transparency code of practice for labeling AI-generated content

Meta said it is signing the EU AI Act Code of Practice on Transparency of AI-Generated Content, a voluntary framework intended to improve how platforms disclose synthetic media in an era of increasingly realistic images and video.

In a statement posted July 28, Meta said it has been building transparency measures since February 2024, when the company announced its approach to identifying and labeling AI-generated content across its platforms. The company framed the effort as a way to give people practical tools to tell what they are seeing, not just comply with rules.

Meta also pointed to work underway beyond labeling alone. It said it recently launched a research demo of a detection tool meant to help people understand whether an image was made using Meta AI.

The company said joining the EU code of practice “reflects our commitment” to transparency as photorealistic AI media becomes harder to distinguish from real content. It also said the decision aligns with similar work with industry peers to tackle what it described as cross-industry technical and operational challenges.

A central theme in Meta’s announcement was caution about how transparency should be presented. Meta said it must provide clarity rather than confusion, and it warned against a scenario where online content ends up covered by “a growing array of different labels and disclosures” that overwhelm users.

Meta’s statement tied the new European step to broader collaboration. It pointed to forums and standards efforts including Partnership on AI and the Coalition for Content Provenance and Authenticity (C2PA), describing those groups as part of an effort to develop identification solutions that are durable, technically-feasible, and sustainable over time.

The company also said it will work with the EU AI Office and other partners to keep the approach practical and “interoperable,” meaning the methods should work across different systems and providers rather than locking transparency into a single platform’s format.

As part of that effort, Meta said it wants the framework to remain genuinely useful for the people the measures are designed to serve, while avoiding additional regulatory complexity for providers.

Meta did not provide details in its announcement about what the EU code will require in concrete terms, such as specific label formats, thresholds for disclosure, or enforcement expectations. It also did not spell out how its own labeling will change as a result of the signature, beyond reaffirming the direction it has already been taking since February 2024.

For users and the industry, the next question is how harmonized the approach becomes across regions and ecosystems. Meta’s involvement in multi-stakeholder standards efforts suggests it expects transparency methods to evolve alongside technical standards, but it remains unclear from the announcement how quickly and how uniformly platforms will converge on shared practices.