New EU AI guidance sets practical expectations for labelling, deepfakes and AI-generated content transparency.
In Brief
Companies are increasingly using AI to create or modify content across marketing, communications and customer-facing channels. As EU transparency obligations under the AI Act move closer to application, this raises practical and operational questions around when AI-generated or AI-manipulated content must be labelled, how disclosures should be made, and whether audiences could reasonably perceive synthetic content as authentic. The draft Guidelines and Code of Practice indicate that regulators will expect companies to understand their AI workflows, assess content-related risks and adopt clear and consistent transparency measures. Preparing early will be important not only to reduce compliance exposure, but also to protect trust, reputation and consistency across markets, as AI-enabled content becomes more embedded in commercial activity.
In more detail
The European Commission has recently published draft Guidelines on the transparency obligations under Article 50 of the AI Act, together with a Code of Practice on the marking and labelling of AI-generated content.
These developments provide the first practical insight into how regulators are likely to interpret and enforce AI-related transparency obligations in real-life scenarios.
With the main transparency obligations expected to apply from 2 August 2026, companies should start assessing their use of AI-generated or AI-manipulated content and adapting their practices accordingly.
A. Guidelines on the transparency obligations under Article 50 of the AI Act
A.1 Deepfake-related disclosure obligations (Article 50(4))
The AI Act introduces specific transparency obligations for certain categories of AI-generated or manipulated content, including the so-called “deepfakes” (Article 50(4)).
The draft Guidelinesprovide helpful clarification and introduce a structured “deepfake test”, based on four cumulative elements:
- an appreciable level of realism or resemblance;
- content that could plausibly exist in the real world;
- reference to real or fictitious persons, objects, places or events; and
- a false appearance of authenticity (i.e. content that may be perceived as genuine by the relevant audience).
From a practical perspective, two aspects are particularly noteworthy:
- First, the requirement of plausibility in the real world does not depend on the depiction of real individuals. It is sufficient that it depicts something that could exist.
- Second, the assessment is audience-based and does not require any intention to mislead. Instead, it focuses on whether the target audience could perceive the content as authentic.
The draft Guidelines further clarify that while minor or purely technical AI edits will generally not qualify as deepfakes; more substantial AI-driven alterations may bring content within scope.
More generally, even where a strict legal obligation may not clearly arise, the regulatory direction points towards increased transparency as a baseline expectation.
A.2 Upstream transparency obligations (Article 50(2))
In parallel, Article 50(2) introduces upstream transparency obligations for providers of AI systems that generate or manipulate content. These systems must ensure that (i) outputs are marked in a machine-readable format and (ii) remain detectable as artificially generated or manipulated.
While these obligations are primarily addressed to AI providers (those developing or placing AI systems on the market) rather than deployers (those using AI systems under their authority in the course of professional activities), they are likely to have indirect implications for the latter, including companies using third-party AI tools.
B. Code of Practice on Transparency of AI-Generated Content
The European Commission has released on 10th June a Code of Practice on the marking and labelling of AI-generated content, which complements the Article 50 Guidelines.
Although currently subject to an adequacy assessment by the Commission and the AI Board, the Code already provides practical, operational guidance for both providers and deployers. In particular, it focuses on:
- Labelling deepfakes and AI-generated or AI-manipulated content in matters of public interest
- Informing users when they are interacting with AI systems (e.g. chatbots)
- Ensuring content can be identified as AI-generated, including through machine-readable marking
The Code offers an early indication of regulatory expectations in practice, and is likely to become a key benchmark for both compliance and trust, especially in customer-facing contexts.
Once formally endorsed, adherence to the Code, companies that choose to adhere to it will be able to rely on its measures as a recognised way to demonstrate compliance with the AI Act’s transparency requirements. This is expected to provide a greater degree of legal certainty and consistency across Member States, as well as streamline implementation from an operational perspective.
Conversely, companies choosing alternative approaches will still need to comply with the AI Act, but may face greater scrutiny and will need to justify and document their chosen measures.
C. Timing of AI Act transparency obligations
The transparency obligations under Article 50 are generally expected to apply from 2 August 2026.
However, under the AI Omnibus agreement approved by the European Parliament in its June plenary session, generative AI systems already on the market before that date will benefit from a transitional period until 2 December 2026 to comply with the machine-readable marking obligation under Article 50(2). This more gradual timeline does not extend to the deepfake-related disclosure obligations, which will become applicable as originally planned (i.e. from 2 August 2026).
Key takeaways for companies
Beyond strict legal compliance, the new EU transparency framework signals a broader shift towards disclosure becoming a standard feature of AI-driven content. The low threshold of the deepfake test combined with its audience-based assessment, means that many everyday commercial practices (particularly in marketing and customer-facing content) may fall within scope sooner than expected.
In this context, companies should start treating transparency not only as a regulatory obligation, but as a strategic decision linked to trust, reputation and future enforcement risk. Early integration of clear and consistent transparency practices will therefore be key to navigating this evolving landscape.
In practical terms, this includes (i) understanding how AI is used across content workflows, (ii) identifying where content may appear authentic to the target audience, and (iii) ensuring that appropriate disclosure measures are applied in a consistent manner. This may also require reviewing internal processes and governance frameworks, including the allocation of responsibilities (e.g. RACI models), to ensure that transparency obligations are implemented consistently across functions. At Baker McKenzie, we would be happy to support you in navigating these developments and designing practical, business-oriented transparency strategies tailored to your needs.
