EU AI Act Article 50 checklist: labelling AI-generated fashion imagery from 2 August 2026

On 2 August 2026, Article 50 of the European Union's Artificial Intelligence Act entered into force, imposing transparency obligations on deployers of general-purpose AI systems that generate synthetic content. For fashion companies using AI to create campaign imagery, product photography, model shots, or social media visuals, the rule is unambiguous: such content must carry both a visible disclosure and a machine-readable watermark when made public in the EU.

The obligation applies regardless of where the company is headquartered. Any brand, retailer, or platform distributing AI-generated or AI-manipulated images to EU consumers must comply or risk fines of up to 1.5 per cent of global annual turnover or EUR 15 million, whichever is higher. For a multinational with EUR 10 billion in sales, that ceiling translates to EUR 150 million.

Three months into the enforcement window, many fashion executives remain uncertain about what qualifies as AI-generated content, which technical standards to adopt, and how to retrofit labelling into existing creative and e-commerce workflows. This checklist provides a step-by-step guide to Article 50 compliance, drawing on regulatory guidance, industry standards, and early adopter experience.

What does Article 50 require fashion companies to do?

Article 50 stipulates that deployers of general-purpose AI models capable of generating or manipulating image, audio, or video content must disclose that the output is artificially generated or manipulated. The provision covers two distinct obligations: a human-readable disclosure, visible to the ordinary viewer, and a machine-readable marker embedded in the file metadata.

The visible disclosure need not follow a prescribed format, but it must be clear, conspicuous, and understandable. Examples include a text overlay reading "AI-generated image" or a small icon in the corner of a photograph. The machine-readable component must conform to a technical standard that enables automated detection by platforms, search engines, and verification tools. The European Commission has indicated that the Coalition for Content Provenance and Authenticity (C2PA) specification, which embeds cryptographically signed metadata in image files, meets this requirement.

The obligation applies to content made publicly available, whether on a brand website, social media account, digital advertisement, or third-party retail platform. Internal design mockups or private mood boards shared within a company's network are not covered, but once an image is published to an audience that includes EU consumers, both labels become mandatory.

Which types of fashion imagery must be labelled?

Article 50 applies to any image that has been wholly generated by an AI system or materially altered using AI tools. Wholly generated content includes fashion campaign photography created from text prompts, virtual model shots, synthetic fabric patterns, and AI-designed product visualisations. Material alterations cover scenarios in which a photograph of a real model or garment is edited using generative AI to change the background, adjust lighting, swap garment colours, or retouch features beyond traditional editing.

The line between routine retouching and material AI manipulation is not yet settled in case law. Industry guidance suggests that minor adjustments using conventional tools such as exposure correction, cropping, or colour grading do not trigger Article 50. By contrast, using a generative fill feature to add or remove objects, change a model's appearance, or create a composite scene does. When in doubt, legal advisers recommend labelling.

User-generated content hosted on brand platforms also falls within scope if the brand exercises editorial control or if the platform itself deploys the AI system that creates the content. A luxury house running a social campaign inviting customers to upload AI-generated outfit ideas, for example, would need to ensure those images carry the required labels if the house provides the AI tool or curates submissions for publication.

How do the visible and machine-readable labels work in practice?

The visible label can take several forms. A short text caption placed near the image, a watermark overlay, or an icon in a consistent location all satisfy the requirement, provided the disclosure is legible and cannot be missed by a viewer scrolling through a product gallery or social feed. The European AI Office has not mandated specific wording, but phrases such as "AI-generated", "Synthetic image", or "Created with artificial intelligence" are widely adopted.

Placement matters. A disclosure buried in small type at the bottom of a webpage footer or hidden behind a tooltip is unlikely to meet the conspicuousness standard. Industry best practice is to place the label adjacent to the image itself, either as an overlay or as a caption line directly beneath the visual.

The machine-readable component relies on the C2PA technical specification, which embeds a content credentials manifest inside the image file. This manifest records the provenance of the asset, including the AI model used, the date of creation, and any subsequent edits. The manifest is cryptographically signed, making tampering detectable. When a user views the image in a compatible browser or application, the software can display a provenance panel showing whether the image is AI-generated.

As of early 2025, the C2PA Steering Committee reported that 47 creative software tools, including Adobe Photoshop, Canva, and Stability AI platforms, supported C2PA metadata embedding, up from 28 in 2024 and 12 in 2023. Adobe, a founding member of the Coalition for Content Provenance and Authenticity, integrated content credentials into its Creative Cloud suite in 2023, enabling designers to attach provenance information automatically when exporting files.

What steps should fashion companies take to comply before deadlines tighten?

1. Audit existing content workflows to identify every touchpoint where AI tools generate or manipulate images. Map the journey from initial prompt or shoot to final publication, noting which software, models, and team members are involved.
2. Adopt C2PA-compatible tools across design, marketing, and e-commerce teams. Ensure that image-editing and generative platforms support embedded content credentials, and train staff to activate these features during export.
3. Establish clear labelling standards in brand guidelines. Specify the exact wording, font size, placement, and visual treatment for visible disclosures, and require all teams to apply them consistently.
4. Integrate labels into digital asset management systems. Configure DAM platforms to automatically append visible watermarks and verify the presence of C2PA metadata before assets are cleared for publication.
5. Train creative and legal teams jointly on what counts as material manipulation. Provide side-by-side examples of compliant and non-compliant edits, and designate a compliance officer within the marketing or legal function.
6. Monitor regulatory updates from the European AI Office and national enforcement authorities. Article 50 is part of a broader regime; future implementing acts may refine technical standards or introduce additional obligations.
7. Prepare for audits and public scrutiny. Document compliance processes, retain metadata logs, and designate a point of contact for regulator enquiries or consumer complaints about unlabelled content.

Which fashion companies have started labelling AI-generated imagery?

Public disclosure of Article 50 compliance efforts remains limited, but a handful of companies have announced initiatives. In March 2025, Adobe published a case study describing how its own marketing team used Content Credentials to label AI-generated campaign visuals. The company reported that embedding C2PA metadata added less than two seconds to the average export workflow and that visible disclosures were placed as semi-transparent overlays in the lower-left corner of each image. Adobe stated that the approach met both the technical and user-experience requirements of the forthcoming EU rules.

Beyond Adobe's self-published example, fashion brands have been quieter. Industry observers note that luxury houses and fast-fashion retailers are testing labelling internally but have not yet issued detailed public statements. At the BoF Voices conference in May 2025, a panel of chief marketing officers from European brands acknowledged that Article 50 compliance was under way but cited competitive sensitivity around the extent of AI use in creative production.

In September 2025, the European Fashion Alliance, a Brussels-based trade association, released a guidance note recommending that members adopt C2PA standards and pilot visible labels on social media channels before the August 2026 deadline. The note did not name specific brands but indicated that several member companies had conducted internal trials during the first half of 2025.

What are the risks and open questions around Article 50 compliance?

Enforcement remains the largest unknown. The AI Act assigns supervisory authority to national competent authorities, but coordination mechanisms are still being finalised. No fashion company has yet been fined for non-compliance, and it is unclear whether regulators will prioritise consumer-facing content over other AI applications in the initial enforcement phase.

Technical limitations also pose challenges. C2PA metadata can be stripped from an image if a user takes a screenshot, re-saves the file in a non-compliant format, or shares the image via a platform that does not preserve metadata. This means that even a fully compliant asset may circulate on social media or third-party sites without its machine-readable label intact. The visible disclosure offers a fallback, but only if it survives cropping and resizing.

Defining the boundary of material manipulation is another grey area. A designer who uses an AI tool to remove a stray thread from a product photograph might argue that the edit is trivial and does not alter the garment's appearance in a way that misleads consumers. Regulators have not yet published detailed criteria, and the first wave of enforcement decisions will clarify where the threshold lies.

There is also debate about consumer perception. Some marketing teams worry that visible AI labels will undermine the aspirational quality of fashion imagery or signal lower production values. Early consumer research is mixed. A survey conducted by the European Consumer Organisation in June 2025 found that 68 per cent of EU respondents wanted clear disclosure of AI-generated advertising, but focus groups revealed that younger consumers were less concerned about synthetic imagery than older cohorts.

How do Article 50 rules compare with other transparency regimes?

The EU AI Act is the first comprehensive legal framework to mandate both visible and machine-readable labels for AI-generated content. Other jurisdictions have introduced disclosure rules, but they vary in scope and stringency. In the United States, California's AB 2655, signed into law in September 2024, requires platforms to label deceptive AI-generated election content but does not extend to commercial or fashion imagery. China's Provisions on the Management of Deep Synthesis Internet Information Services, which took effect in January 2023, require platforms to mark AI-generated content with a visible label but do not prescribe a machine-readable standard.

The United Kingdom has taken a voluntary approach. In February 2025, the UK's Department for Science, Innovation and Technology published a code of practice encouraging deployers of generative AI to adopt C2PA metadata, but compliance is not legally binding. Fashion companies selling in both the EU and UK markets are therefore subject to different rules, with the EU regime imposing hard obligations and the UK relying on industry self-regulation.

Multinational fashion groups must therefore manage a patchwork of labelling requirements. A global campaign using AI-generated models will need visible and machine-readable labels for EU distribution, visible labels only for China, and no mandatory disclosure in the United States outside election contexts. Digital asset management systems must be capable of versioning assets by region and applying the appropriate metadata and overlays.

What should fashion companies do now?

1. Audit content libraries immediately. Identify all AI-generated or AI-manipulated images currently in use on EU-facing websites, social channels, and retail platforms, and prioritise labelling for high-traffic assets.
2. Upgrade creative software stacks. Ensure that all design, photography, and marketing tools support C2PA metadata embedding, and phase out legacy systems that cannot attach content credentials.
3. Draft and publish internal guidelines. Codify the definitions of AI-generated and materially manipulated content, specify labelling formats, and assign accountability to named roles within marketing and legal teams.
4. Integrate compliance into DAM workflows. Configure digital asset management platforms to block publication of unlabelled AI content, and require metadata validation before final approval.
5. Train cross-functional teams. Run joint workshops for designers, marketers, and legal counsel to align on what triggers Article 50, how to apply labels, and how to document compliance.
6. Monitor enforcement developments. Subscribe to updates from the European AI Office, national data protection authorities, and industry associations to catch implementing acts, guidance notes, and early enforcement cases.
7. Prepare for public transparency. Develop a consumer-facing FAQ explaining how the brand uses AI in creative production, why images are labelled, and what protections are in place to ensure accuracy and authenticity.

Frequently asked questions