Starting Sunday, August 2, 2026, the European Union will require companies to label artificially generated content that is designed to look authentic (The Guardian). The rules apply to new AI systems entering the EU market from that date. Existing systems get a four-month extension to comply.

The obligations come from Article 50 of Regulation (EU) 2024/1689, the EU's AI Act, which the European Commission calls the first comprehensive legal framework on AI worldwide (European Commission). The Act was first agreed in December 2023. Ahead of the August 2 deadline, the Commission published guidelines on transparency obligations for providers under Article 50 on July 29, 2026 (European Commission), following a Code of Practice on marking and labelling AI-generated content released June 10, 2026 (European Commission). A second draft of that Code had been published March 5, 2026 (European Commission).

Synthetic text, images, video, and audio designed to look truthful must carry a visible mark indicating AI generation and contain a digital watermark showing artificial origins. Think of the visible mark as a nutrition label on food packaging — something consumers can see at a glance — while the digital watermark works more like a serial number embedded inside the product, machine-readable but invisible to the eye. Text on matters of public interest must also be labeled as AI-made if produced without human editorial oversight. The rules do not apply to personal user content. There is an exemption for evidently artistic, satirical, and fictional works. Companies are encouraged, though not required, to label AI-generated content predating the regulations.

The European Commission has created black-and-white AI labels available for anyone to use, though companies may design their own. Failure to comply can result in fines of up to €15 million or 3% of a company's worldwide global turnover.

Green MEP Sergey Lagodinsky, who helped negotiate the AI Act, said the transparency rules are a matter of customer protection and democracy protection.

The broader context here involves a dispute over the scope of these transparency obligations. The Computer and Communications Industry Association (CCIA Europe) argues that EU guidelines published in July 2026 expanded the definition of a deep fake beyond what was established in the 2024 AI Act. Boniface de Champris, CCIA Europe's AI policy lead, represents the industry group's position on the matter.

This disagreement touches on the foundational legal architecture of the AI Act itself. The Commission's August 2025 guidelines on prohibited artificial intelligence practices, hosted on its AI Act Service Desk domain, laid additional groundwork for how the regulation's provisions are interpreted and enforced (European Commission). The progression from the original 2024 legislative text to the July 2026 guidelines illustrates the ongoing interpretive process that regulated entities must navigate.

For affected firms, the distinction between new and existing AI systems creates a tiered implementation schedule. Companies launching new AI products on the EU market from August 2, 2026 face immediate compliance requirements. Those maintaining pre-existing systems have until late 2026 to integrate visible markings and digital watermarks into their synthetic content pipelines. The design flexibility offered by the Commission's black-and-white label template allows entities to adopt a standardized approach immediately or develop proprietary labeling systems, provided the core visibility and watermarking requirements are met.

In my view, the tension between the Commission's expanded deep fake definition and industry's reading of the original 2024 legislative text presents a live regulatory question. How enforcement authorities interpret the boundary between authentic-looking synthetic content and the exemptions for artistic or satirical works will shape compliance strategies across the digital single market. The severity of the financial penalties, reaching up to 3% of global turnover, gives regulators substantial leverage in any dispute over labeling sufficiency or definitional scope.

The implementation date arrives roughly two and a half years after the initial political agreement on the AI Act in December 2023. The staggered timeline, with existing systems receiving additional months to comply, acknowledges the technical challenge of retroactively applying digital watermarks to deployed AI systems.