What happened
On 2nd August, the transparency obligations under Article 50 of the EU AI Act came into effect, backed by guidelines from the European Commission. Most coverage treated this as another compliance deadline: label your AI content or risk a fine.
The more hidden details are more useful: the Commission’s guidance draws a line between AI-generated content that needs disclosing and text that doesn’t – that line runs through the editorial process, not through whether AI touched the work at all.
For brands with a disciplined, agency-led content operation, this rule change looks less like a new burden and more like a formal description of what’s already in place.
What actually changed
Article 50 sets out several overlapping duties. Two matter most for marketing teams, according to the Commission’s guidance:
- Article 50(4), the Commission’s guidance states that AI tools are required to disclose when a deep fake is created or used. But what precisely is meant by ‘deep fake’? The Commission’s guidance describes it as covering any AI-generated or AI-manipulated image, audio or video that resembles a real person, object, place or event closely enough to appear authentic when it isn’t. The guidance also extends this to AI-generated text used to inform the public on matters of public interest, a category the Commission says can include certain advertising, PR and corporate communications, though where exactly that line sits is a legal judgement rather than a marketing one.
- Article 50(2) requires machine-readable marking of synthetic content more broadly. Generative systems already on the market before 2nd August reportedly have until December 2026 to meet this specific marking requirement.
The guidance also points to an exemption for text: disclosure appears not to be required where AI-generated copy goes through meaningful human review, with a named human editor or editorial entity taking ultimate responsibility for the content. Creative work that’s clearly artistic, satirical or fictional is treated more lightly too, though the guidance is clear that commercial content rarely qualifies once it mixes creative and promotional intent.
The obligation isn’t tied to where a company is based. Non-EU companies can fall under this jurisdiction if their AI-generated content reaches EU audiences, according to the Commission’s guidance, though the precise boundaries of that exposure are worth confirming with your legal advisors.
Why this should be on your agenda
Reach determines exposure
The Commission’s guidance indicates that a non-EU advertiser producing AI-generated content for marketing material that runs in the EU can fall under this scope. For brands running pan-European or global activity from outside the EU: what seems to matter is where the content lands, not where it was made, though the exact boundary is one to confirm.
The definition of “deep fake” is wider
Based on the Commission’s own examples, an AI-generated product shot that makes an item look better than the real thing, would likely count. A synthetic influencer testing a sponsored product would likely count too. An AI-generated background behind a real, unaltered product generally wouldn’t, because it isn’t likely to mislead anyone about what’s actually being sold. The guiding idea appears to be whether content could pass as authentic and change how someone assesses it.
The editorial process point is the one worth sitting with
This is the part that rewards structure over speed. The Commission’s guidance points to human review and named editorial accountability, not the mere involvement of AI as the deciding factor for AI-assisted copy. Read plainly, that looks like a description of what a properly run, AI-supported content operation already does: AI drafts, a specialist edits for tone of voice and to ensure accuracy, then a named person signs off before anything goes live. Whether a specific piece of copy meets the legal bar is an open question each brand needs to decide for themselves, but a common thread appears to be that a stringent, human editorial process is rewarded.
Multilingual campaigns add more points to check
Each market version of a campaign, translated, localised and often re-rendered, is a separate asset to consider under the Act. The guidance suggests disclosure needs to be clear at first exposure in every market it appears, not just the originating one. Worth checking whether your review process already treats every market version this way.
What this means for your content operation
- Audit your AI-generated creative library, market by market. Product shots, synthetic backgrounds and any AI-generated human likeness are worth a second look.
- Make sure AI-assisted copy has a named reviewer. A general assurance that “AI output gets checked” is weaker than a specific person who owns the sign-off and it’s a better content practice regardless of what the Act ultimately requires.
- Build disclosure decisions into the localisation step. Each market version of a campaign is handled separately in production; treating disclosure the same way avoids it becoming an afterthought once translation is done.
- Check with legal before the next major AI-assisted marketing goes live, particularly for anything reaching EU audiences. This is a compliance question with real financial stakes and it deserves a proper legal read.
Brands that already route AI-assisted content through a specialist for fact-checking and sign-off have less to change than the headlines suggest. The bigger shift is for brands publishing AI output straight through with no review step at all.
Sources
- European Commission, Article 50 of the AI Act and Commission Guidelines
- European Commission, “Commission starts enforcing AI Act rules and new transparency requirements on 2 August”
- Davis+Gilbert LLP, “EU AI Act Guidance Expands AI Disclosure Rules for Advertisers and PR Teams” (via Mondaq)
- Stibbe, “The AI Act’s Transparency Obligations: Rules, Scope and Timeline”
Article author
