European Commission issues voluntary Code for labelling AI content

The European Commission has published a voluntary Code of Practice to help providers and deployers mark and label AI‑generated or AI‑manipulated text, audio, images and video.

The Code is intended to help organisations meet transparency obligations under the EU AI Act, which come into force on 2 August 2026, and offers practical recommendations on visible and machine‑readable labelling to aid compliance, the Commission says.

The Commission frames the Code as guidance for both creators of generative systems and the organisations that deploy them — explicitly naming deployers such as platforms, newsrooms and employers as intended users. It sets out principles and technical options for signalling when content is synthetic or has been altered by AI, and recommends accompanying metadata and provenance information to support automated detection and consumer understanding.

For HR leaders and technology teams, the guidance is pitched as practical: the Commission points to measures that can be adopted during content production and distribution, and it highlights machine‑readable labelling formats so downstream systems can identify synthetic assets. The Code also stresses the importance of transparency for trust and legal compliance, positioning labelling as one element of broader obligations under the AI Act’s transparency regime.

The move follows a wave of regulatory attention to generative models and manipulated media. Across Europe and beyond, regulators and lawmakers have pressed for clearer signals when content is produced or substantially altered by AI, citing risks from misinformation, privacy intrusions and unfair automated decision‑making. The Commission’s Code aligns with those aims while remaining voluntary, offering a blueprint that could be referenced by industry and national authorities as they prepare for the AI Act’s enforcement date.

The Commission acknowledges trade‑offs in its document: visible labels help human audiences, while machine‑readable metadata supports automated compliance and content provenance systems. It also recognises different technical approaches — from embedded metadata to watermarking and external registries — and suggests deployers weigh robustness, usability and interoperability in choosing methods.

What the Code does not do is create binding obligations. The Commission is explicit that the document is voluntary and stops short of prescribing a single technical standard or mandating certification of labelling tools. It does not set timelines for adoption, nor does it establish an independent audit or approval process for labelling methods. The Code also leaves open how conflicts between labelling duties and other legal requirements — for example, intellectual property, trade secrets or data‑protection constraints in workplace settings — should be resolved in practice.

For employers already experimenting with generative AI, the Commission’s guidance signals that transparency requirements are moving from theoretical to operational priorities. HR and legal teams will need to map where AI is used to create or alter content, consider how labelling is implemented in employee‑facing tools, and prepare policies that reflect the difference between voluntary guidance and the AI Act’s enforceable duties after August 2026.

The Code is likely to shape vendor and platform responses in the months ahead: some providers may adopt voluntary labelling features to help customers demonstrate compliance, while others could wait for harmonised technical standards or national guidance. Either way, the Commission’s publication marks a clear nudge toward routine disclosure of AI involvement in workplace communications and public‑facing content — a shift that will affect procurement, policies and everyday practice as organisations ready themselves for the AI Act’s transparency regime.