AI transparency and deepfakes in marketing: new European rules from August 2026

The 4 conditions of the new European deepfake test

The AI Act introduces specific obligations for certain categories of AI content, including the widely discussed deepfakes, enshrined in Article 50(4). The new draft guidelines clarify this by introducing a structured deepfake test based on four cumulative elements evaluated in a sequential assessment:

• There must be a significant level of realism or resemblance.
• The content must be capable of plausibly existing in the real world.
• It must reference real or fictional persons, objects, places, or events.
• Finally, a false impression of authenticity is required, whereby the relevant target audience could perceive the content as genuine.

Plausibility and perception: when is AI content ‘authentic’?

From a legal and practical point of view, two elements are crucial in this assessment. First, the requirement of plausibility in the real world does not depend on depicting existing individuals. It is sufficient that the image shows something that could actually exist.

Second, the assessment is based purely on the audience and requires no intention to deceive. The focus rests solely on whether your target audience could perceive the content as authentic. This highly subjective requirement will undoubtedly spark intense debate regarding consumer perception during legal disputes and regulatory enforcement. The guidelines do clarify that minor or purely technical AI edits generally remain exempt, whereas more substantial modifications using AI irrevocably bring the content under these strict rules.

Upstream obligations for providers of AI systems

In addition to the rules for the content itself, Article 50(2) introduces so-called upstream transparency obligations for providers of AI systems. They must guarantee that the output is marked in a machine-readable format and remains detectable as artificially generated. Although these rules are primarily directed at the developers placing these systems on the market, this has indirect yet undeniable consequences for you as a user of such third-party tools in a professional context.

To make this workable in practice, a new Code of Practice offers concrete operational guidance. The focus here lies on labeling deepfakes in matters of public interest, informing consumers when they interact with AI systems such as chatbots, and ensuring that files are identifiable as AI-generated through machine-readable markings. Once this code is formally approved, your company can use these guidelines as a recognized method to demonstrate compliance with the AI Act. This provides much-needed legal certainty throughout the European Union. Businesses that prefer an alternative approach must, of course, still comply with the legislation. However, they risk significantly stricter oversight, requiring them to independently justify and meticulously document their choices.

The clock is ticking: crucial deadlines in August and December 2026

It is now July 2026, and the time for waiting is definitively over. The general transparency obligations under Article 50 in principle take effect as early as 2 August 2026. That is literally next month. Under the AI Omnibus agreement, generative AI systems placed on the market before that date receive a brief breathing space until 2 December 2026 to comply with the requirement regarding machine-readable markings. Make no mistake regarding this apparent leniency, however. This phased transition absolutely does not apply to the disclosure obligation for deepfakes. That obligation takes effect on 2 August without any delay and in its full asymmetric weight?

What does this mean in concrete terms for your company?

The new European parameters make it painfully clear that transparency regarding AI content is rapidly becoming the absolute standard in business. The particularly low threshold of the deepfake test, coupled with the assessment based on public perception, means that many everyday commercial and marketing practices are already coming under scrutiny today.

You should not view these transparency rules merely as a mandatory checkmark on a compliance checklist. It is a strategic imperative to maintain customer trust and eliminate acute enforcement risks. In practice, this requires you to immediately map out how artificial intelligence flows through your entire content workflows. You must critically analyze where certain content could be perceived as authentic by your target audience and ensure that appropriate disclaimers are applied consistently and visibly. This demands a fundamental review of your internal governance, including a clear assignment of responsibilities using RACI models.