MetaがEU AI法「AIが生成したコンテンツの透明性に関する行動規範」に署名Meta is Signing the EU AI Act Code of Practice on Transparency of AI-Generated Content

Meta has signed the European Union's AI Act Code of Practice on the transparency of AI-generated content, formally committing to disclosure and labeling standards for material produced or altered by its artificial intelligence systems. The move matters because it aligns one of the world's largest platform operators with the transparency provisions of the EU AI Act at a moment when those rules are moving from principle to enforcement, and it offers a signal of how major technology firms intend to demonstrate compliance in the bloc.

The code sits alongside the broader legal obligations set out in the AI Act, the EU's comprehensive framework for regulating artificial intelligence, which entered into force in 2024 and applies in phases. Transparency requirements are a central pillar of that law. Under Article 50, providers of generative systems are expected to mark synthetic audio, image, video, and text in a machine-readable format so it can be detected as artificially generated, while deployers are expected to disclose when content such as deepfakes has been created or manipulated by AI. A code of practice is a voluntary instrument that translates those legal duties into more concrete, agreed-upon commitments, giving signatories a recognized route to show they are acting in good faith while the regulatory details are worked out.

For Meta, the commitment builds on labeling efforts the company has already introduced across Facebook, Instagram, and Threads, where content identified as AI-generated has been tagged with markers indicating its origin. Meta has also participated in industry work on provenance, and the practical implementation of transparency rules increasingly relies on shared technical standards. These include content credentials developed under the Coalition for Content Provenance and Authenticity, known as C2PA, which embed tamper-evident metadata describing how a piece of media was made, as well as watermarking approaches that encode signals directly into generated media. Detecting AI content reliably remains difficult, however, because metadata can be stripped during uploading or editing and watermarks vary in robustness, so the effectiveness of any labeling regime depends heavily on cross-industry adoption.

The signing appears to be part of a wider pattern of technology companies engaging with the EU's voluntary instruments rather than resisting them outright. The AI Act framework has been accompanied by separate codes addressing general-purpose AI models, covering areas such as safety, copyright, and transparency, and several large developers have weighed whether to endorse them. Meta's stance toward EU AI governance has not been uniform; the company has previously expressed reservations about aspects of the bloc's approach and about the pace and clarity of some rules. Its decision to sign this particular content-transparency code therefore suggests a selective engagement, endorsing commitments tied closely to labeling and disclosure while broader debates about model-level regulation continue.

The context extends beyond Europe. Regulators and lawmakers in other jurisdictions, including parts of the United States and Asia, have been examining disclosure requirements for synthetic media, often driven by concerns about election-related deepfakes, fraud, and non-consensual imagery. Because platforms operate globally, standards adopted to satisfy EU obligations are likely to influence practices in other markets, a dynamic sometimes described as the "Brussels effect." Standard-setting bodies and industry groups continue to refine provenance specifications, and interoperability between different watermarking and metadata systems remains an open technical and governance question.

The practical impact of the commitment will depend on implementation. Signing a code establishes intent and a set of reference expectations, but the value to users rests on whether labels are applied consistently, whether they survive as content circulates, and whether disclosure is clear enough to be understood without being intrusive. It also remains to be seen how voluntary codes will interact with the binding obligations that take effect under the AI Act, and how regulators will treat adherence to a code when assessing compliance.

For now, the announcement represents a compliance milestone and a public alignment with the EU's transparency agenda. Whether it meaningfully improves users' ability to distinguish AI-generated material from authentic content will become clearer as the labeling systems are deployed at scale and as other providers, platforms, and tool makers decide how closely to coordinate their own approaches.