The EU AI Act reaches creators through Article 50, which the European Commission's FAQ says has applied since 2 August 2026. If you use AI tools as part of a business rather than purely for personal fun, the Act treats you as a deployer, and you must tell people when an image, audio clip or video you publish is a deepfake, and when AI-written text you publish to inform the public on matters of public interest has not been through human review with someone holding editorial responsibility. Marking outputs in machine-readable form is a separate duty that falls on the providers of the AI tools, which for most creators means the companies behind them rather than you.
Are you a deployer? The professional-use test
The Act splits responsibility between providers, who develop AI systems and put them on the market under their own name, and deployers, who use those systems under their own authority. The Commission's FAQ excludes personal, non-professional use, and gives the example of someone generating deepfakes and sharing them on social media in a personal capacity. The line moves as soon as money is involved: if you gain an economic benefit from the activity on a regular basis, or it is part of a business, trade, occupation or freelance work, you are a deployer of the tools you use. For anyone earning from their content, that describes the normal situation.
The same FAQ settles a question agencies ask. When a company uses an AI system, the company is the deployer, and the employees who operate it, such as content creators or digital animators working under its instructions, are not separate deployers. The company stays the deployer even when contractors or freelancers run the tool on its behalf and under its control.
Location is not a safe harbour either. Article 2 of the Act extends it to providers and deployers established outside the EU where the output produced by the AI system is used in the Union. A creator based in Australia, the UK or the US with EU followers should therefore treat EU-facing posts as in scope; how a national authority would pursue someone abroad is not something the guidance addresses.
Decision tree: does Article 50 apply to this post?
Run each piece of content through these questions in order. The wording follows the operative text on the AI Act Service Desk's Article 50 page and the Commission's FAQ.
- Is this purely personal, non-professional use? If yes, the deployer duties do not apply. If you earn from your content or post as part of a business, keep going.
- Is the item an image, audio clip or video made or changed with AI? If not, jump to question five.
- Does it resemble a person, object, place, entity or event that exists, or could plausibly exist? If not, it is not a deepfake under the Act's definition.
- Would it falsely appear authentic or truthful to the audience you are posting it to? If yes, it is a deepfake and needs a clear disclosure no later than first exposure. If it forms part of an evidently artistic, creative, satirical or fictional work, the disclosure can be lighter, provided it is made in a way that does not spoil the work.
- Is it text you publish to inform the public on a matter of public interest? If yes, label it unless a person with relevant knowledge reviewed its substance and someone holds editorial responsibility for publishing it. Spell-checks and grammar fixes do not count as review.
- Are you offering an AI system that talks directly with people, such as a chat persona? The duty to tell users they are dealing with AI sits with the provider of that system. If you built it, or put it into service under your own name, get advice on whether that makes you the provider.
- Does a tool you use read emotions or sort people by biometric traits? Then the people exposed to it must be informed. Few creators use these, but some interactive tools do.
Why photorealistic personas usually count as deepfakes
The FAQ sets three cumulative criteria for a deepfake: a high level of resemblance to the simulated subject, a subject that exists or could plausibly have existed, and a false appearance of being authentic or truthful. The second criterion catches many creators by surprise. A photorealistic AI influencer who has never existed can still meet it, because a lifelike person who could plausibly exist is enough.
That leaves the third criterion, which turns on context. The Commission's guidance lets deployers weigh the level of resemblance, the substance of the message, the deployment context and what the intended audience expects. If followers believe a persona is a real human, its content falsely appears authentic, which is exactly what disclosure is meant to correct. A clearly stylised anime character is much less likely to meet the resemblance tests at all, and the FAQ gives background scenes and special effects in standard film production as examples unlikely to mislead an audience. Because disclosure must reach people at first exposure, a statement in your bio may not be enough for someone who meets a clip reshared out of context.
Dated labelling checklist
| Date or trigger | What it means for a creator | Source |
|---|---|---|
| 2 August 2026 | Article 50 applies, including the deployer duties to disclose deepfakes and unreviewed public-interest AI text | Commission Article 50 FAQ |
| Content made before 2 August 2026 | No duty to label it retroactively, although the Commission encourages deployers to do so where possible | Commission Article 50 FAQ |
| 2 December 2026 | Ends the limited grace period for providers of systems already on the market to add machine-readable marking; deployer duties were not deferred | Commission Article 50 FAQ |
| Every new post | Disclosure must be clear and distinguishable no later than the first exposure, and meet accessibility requirements | Article 50(5), AI Act Service Desk |
| Reshares and downloads | An EU icon, if you use one, should remain visible when the content is reshared or downloaded | EU icons for labelling AI content |
Turn those dates into habits with this pre-publication list:
- Place the disclosure on the content itself, as on-screen text, a spoken line or an embedded icon, rather than only in metadata or a bio.
- Use plain words such as “AI-generated” or “voice created with AI”, and avoid abbreviations other than AI.
- For satire, fiction or art, pick a disclosure that leaves the work intact, such as an end card or caption line, but never drop it.
- Carry the disclosure through every cut-down, clip and cross-post of the same asset.
- Have a knowledgeable person review AI-written informational text on public-interest topics before it goes out, and note who did it.
- Keep a simple log of which tool generated or altered each asset and when.
- Do not count a platform's automatic label as your disclosure; the FAQ says deployers cannot rely on providers' machine-readable marks alone.
The Code of Practice and the EU icons
The Code of Practice on Transparency of AI-generated Content was drawn up by independent experts in a process run through the AI Office, and the Commission and the AI Board have confirmed it is an adequate voluntary tool for showing compliance. It has a providers section on marking and detection and a deployers section on labelling deepfakes and AI-generated text. Signing is voluntary, but the obligations are not: deployers who do not sign must show that their own measures are adequate, and the FAQ warns they may face more requests for information.
The EU has also published a free set of icons for deployers, including creators. There is a basic icon for any AI involvement, a variant for content fully generated by AI and one for content that was partially modified. The icons page says they are optional, do not establish compliance by themselves, and may be used without attribution, though using them is not a signal that you have signed the Code. Its display rules ask for an icon that is visible at first exposure, free of overlapping interface elements, embedded in the content unless an equivalent overlay exists, readable by assistive technology, and still visible after a reshare or download. The page also reports that in user testing the icon worked better alongside a short text label, so pairing the two is the sensible default.
Where platform labels and chat personas fit
TikTok's AI-generated content setting and Instagram's Add AI label toggle both put a visible notice on a post, which makes them a practical part of an Article 50 routine. They are platform features rather than EU compliance tools, though, so check that the label actually appears where a first-time viewer will see it. Meta, for example, says that content its systems detect as only modified with AI gets its label in the post menu rather than on the post itself, which may not be prominent enough for a face swap on real footage. Our guides to Instagram AI labels and the TikTok AI-generated content label cover the settings and the automatic labels in detail.
AI chat personas raise a second set of rules that come from platforms rather than the Act. Fansly prohibits AI that interacts with users without human oversight, and Patreon bars AI-driven adult chatbots that can generate content, so check the platform position in our AI persona platform matrix before building one.
Who enforces Article 50
National market surveillance authorities do most of the enforcing. The AI Office has a limited role covering systems built on general-purpose AI models by the same company, or integrated into very large online platforms and search engines, and the European Data Protection Supervisor covers EU institutions. According to the Commission's FAQ, fines can reach 15 million euros or 3% of total worldwide turnover for the preceding financial year, with proportionality taken into account for small and medium-sized businesses. Article 50 also sits alongside other EU and national transparency rules rather than replacing them, so advertising and consumer law still apply to sponsored AI content.
Limitations of this guide
The Commission's FAQ and guidelines are its interpretation of the Act, not the final word; national authorities apply them and courts can read the text differently. The FAQ page used here was last updated on 24 July 2026 and the guidelines page on 6 August 2026, and both may change as enforcement practice develops. This guide does not cover GDPR, consumer law or national rules on image rights, all of which can apply to the same post. It is general information rather than legal advice, so speak to a lawyer familiar with EU technology law before relying on it for client work or AI chat products. For the non-EU parts of the picture, see the AI creator compliance checklist.