Article 50 if you are not in the EU
Duty follows the audience, not your country. A US creator with EU viewers still labels deepfakes. Pre-2 Aug published files are not retroactive.
Article 50 follows the audience, not the creator's passport. A US, UK, or Indian creator who posts a photoreal deepfake that EU viewers can see is in the territorial scope the guidelines describe, even if the laptop never left Brooklyn. Content generated and published before 2 August 2026 is not retroactive. Content you are posting now is. This page is that map. It is not a second tour of the four duties.
The Commission's FAQ (last updated 24 July 2026) is the source for extra-EU providers whose output is used in the EU, for professional versus personal use, and for non-retroactivity of content generated prior to 2 August 2026. Bird & Bird's insight on the guidelines adopted 20 July 2026 is the source for deployer territorial scope, including posts on the globally accessible internet. Paul, Weiss (4 August 2026) is the source for the generation-versus-publication cut on Article 50(4). Morgan Lewis (12 August 2026) confirms deployer duties have applied since 2 August with no grace.
The audience is the trigger
The FAQ on providers: providers established or located outside the EU are subject to the AI Act if the output of their AI system is used in the EU. That sentence is aimed at people who place a system on the market. Most creators are not that person. They prompt a hosted generator and publish the file. They are deployers.
Bird & Bird on deployers: the obligations under Article 50(3) (emotion recognition or biometric categorisation) and Article 50(4) (deepfakes and certain public-interest text) arise both when the deployer is located in the EU and when the deployer is outside the EU and the output produced by the AI system is used in the EU. The final guidelines treat "use in the EU" as happening where the deployer itself foresees dissemination and use of the AI outputs in the EU — for example by directing or authorising distribution within the EU. They also suggest that this happens when a deployer posts deepfakes "on the globally accessible internet," without a requirement that the platform or the content specifically target the EU market.
That last clause is the one a US creator actually lives in. TikTok, YouTube, Instagram, and X are globally accessible. A public For You post is not a mailing to one state. Geo-blocking the EU is the only clean product answer Bird & Bird flags for shielding online content from Article 50(4). A bio that says "based in Austin" is not geo-blocking.
Article 50 is the name of the rule. Synthetic media disclosure is the broader obligation. Neither one asks for an EU VAT number before it attaches.
What "used in the EU" is not, in these sources:
- It is not "you sold ads in Germany." Organic reach counts as dissemination if the file is there to be watched.
- It is not "you geo-targeted a media plan at France." Paid targeting is one way to foresee EU use. It is not the only way.
- It is not "a single EU viewer found you by accident, therefore you are a provider." Incidental, unforeseeable, or unauthorised downstream use, Bird & Bird records, should not alone trigger the provider obligations on a third-country system vendor. That caveat is written for providers. Deployers who chose a public global post are in a different sentence.
If you run a campaign that is geo-fenced out of the EU and the organic post is also blocked there, you have a territorial story. If you run a public account and hope EU viewers will not find it, you do not.
Professional use, not a passport stamp
Territorial scope still sits on top of the deployer definition. The FAQ: when a natural person uses an AI system in their personal capacity — for example to generate deepfakes and disseminate them on social media — that is a personal activity, excluded from the Act. If it is an activity through which they gain an economic benefit on a regular basis, or they are otherwise involved in a business, trade, occupational, or freelance activity, they are a deployer.
That test does not mention nationality. A paid US creator is in the professional column. A hobbyist posting for no regular economic benefit is in the personal column, wherever they live. The kitchen table is not the exit; the absence of regular economic benefit is.
Employees of a non-EU company are not separate deployers. The legal person is. If the brand is Delaware-incorporated and the ads run on Instagram worldwide, the brand is the deployer of the system it used, and EU viewers are the territorial hook. Bird & Bird's agency reading still applies: if a US brand commissions a US agency and does not control whether or how AI is used, the agency holds the labelling duty. The audience does not move because both signatures are American.
Fines, if they ever attach, cap at €15 million or 3 percent of worldwide turnover. Article 50 fines is that page. Worldwide is worldwide. The FAQ does not haircut the cap to EU revenue. Do not plan as if a US tax return is a shield. Also do not invent an enforcement case. The sources on this page state the ceiling and the territorial rule. They do not attach a docket to a particular US creator.
Platform policy is a separate stack and it does not care where you live either. TikTok's Creativity Program, YouTube's altered-content disclosure, and Meta's C2PA/self-disclosure path attach to the login. SocialScale Hub's 2026 recap and Vexub's 2026 monetization recap are the dated platform sources here. You can be outside Article 50's professional-use definition and still lose a program for undisclosed AI. You can satisfy every platform toggle and still owe a perceivable first-exposure mark to an EU viewer. Stack them.
Globally accessible posts and geo-blocking
Bird & Bird is candid that the broad view of "used in the EU" makes it hard to shield online content from Article 50(4) unless some form of geo-blocking can be applied for the EU market. That is the operational fork.
Public global post. Assume EU viewers. If the cut is a deepfake under the FAQ's three-part test — resemblance, existing (including plausibly existing) subject, false appearance of authenticity — put a visible or audible label at first exposure. The decision tree is the walk. The disclosure template is the copy. Do not wait for a geo-report in Studio before you burn the line.
EU-blocked distribution. If you can actually stop the file at the EU border — a geo-gated campaign, a product page that does not resolve in the Union, a platform control that holds — the territorial story changes. "We turned off Europe in ads manager" does not block the organic Reel you posted the same afternoon. Check the organic path. Check embeds. Check downloads that someone in the EU can still open.
Directed EU distribution. A launch that names Member States, a creator code for a European shop, a paid package aimed at DE/FR/ES: you have directed use in the EU. The guidelines' "foresees dissemination" example is this row. Label before the first euro of spend.
Providers outside the EU who place a generator on the EU market, or whose output is used in the EU, have their own marking clock. A US creator prompting a hosted model is usually not that provider. You still cannot hide behind a missing watermark. The FAQ is explicit that deployers cannot simply rely on the provider's machine-readable mark to fulfil the disclosure a person has to perceive.
The Article 50 timeline is the dated calendar: 20 July 2026 guidelines, 2 August 2026 application, 2 December 2026 provider marking grace, 2 February 2027 watermark-detection interoperability. None of those dates move because the laptop is in California. 2 August already applied. The later two dates are provider dates. They are not a US-creator holiday.
The 2 August cut for old files
The FAQ (24 July 2026): content generated prior to 2 August 2026 does not need to be labelled retroactively. The Commission encourages relevant deployers to label older published work where possible. It does not require you to walk the archive.
Paul, Weiss (4 August 2026): content generated — or, in the case of Article 50(4), published — prior to 2 August 2026 does not need to be retroactively labelled.
The working rule this site uses: content generated and published before 2 August 2026 is not retroactive. Generated before but published after is covered. A July render on a US drive, posted in September to a global TikTok, is a September post. The generation country does not matter. The publication date does.
Bird & Bird records a nuance in the final guidelines: for image, audio, and video deepfakes, the relevant date under that reading is the date of generation; for public-interest text, the relevant date is the date of publication. If you are holding unreleased audiovisual takes from July, the conservative move is still to label them on a post-2 August upload. If you are holding unreleased public-interest text, the FAQ's publication cut is clean: label or review before you hit publish.
Do not relabel a 2025 archive because you found this page. Do label the next photoreal talking head that will be watchable in the Union.
If you generate that talking head in Versely's AI video generator, the export does not know where the viewer lives. Burn the first-exposure line into the file so a download in Madrid still carries it. Platform toggles remain on top. If you need a studio seat to put the words on the picture, sign up.
A US-only audience setting that the platform does not actually enforce is not a territorial plan. Check the control. Then check a VPN. Then label anyway if the control is theatre.
FAQ
I have a few EU viewers by accident. Does that count?
Bird & Bird's "incidental, unforeseeable or unauthorised" caveat is written for third-country providers, not as a free pass for a public deployer post. A public video on a globally accessible app is the guidelines' example of EU use. A private file forwarded against your instructions is closer to unauthorised downstream use. If the account is public, label the deepfake.
Does a US-only TikTok setting remove the duty?
Only if it actually stops EU access. The guidelines' shield is geo-blocking, not a settings label. If EU viewers can still open the post, the audience is still there. Paid geo-targeting off Europe does not block the organic path.
Do I have to relabel my 2025 archive?
No. Generated and published before 2 August 2026 is not retroactive. The Commission encourages voluntary labelling of older published work; it does not require an archive rebuild. Anything you publish now, including old renders, is on the current clock.