On 2 August 2026, the transparency rules in Article 50 of the EU AI Act started to apply. If your company sells into Europe and runs video, images or audio that were generated with AI, the duty to say so is now a legal obligation rather than a best practice. Breaches carry fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher.
Most of the coverage has treated this as a problem for AI labs. It is not. A large part of Article 50 lands on the company that publishes the content.
The two sides of the rule
Article 50 splits the work between providers and deployers.
Providers, meaning the companies that build and supply the generative systems, have to mark their outputs in a machine-readable way so the content is detectable as artificially generated. Systems already on the market when the rules took effect have until 2 December 2026 to comply with that marking requirement.
Deployers, meaning the companies that use those systems under their own authority, have their own duty: when a deployer uses AI to create a deepfake, it has to disclose that the content is artificially generated. The EU defines a deepfake broadly enough to cover synthetic video of people, places and events that a viewer could take for real footage. A product film of a machine that exists, shot in a plant that exists, generated rather than filmed, sits inside that definition. A dragon does not: clearly fantastical or physically impossible content falls outside it.
The obligation follows the decision. Whoever decides that AI is used, and how, is the party the regulation is aimed at.
Who counts as the deployer, you or your agency
This is the part worth reading twice, because it decides whose name is on the exposure.
A company remains the deployer when a third party operates the system on its behalf, as long as that third party acts under the company's responsibility and control. Where the contractor is free to decide whether and how to use AI on the assignment, the contractor becomes the deployer.
In practice, that means a brand that hands a production company a brief saying "make it look cinematic, we don't care how" has quietly moved the legal role onto the supplier, and the supplier may not know it. A brand that specifies the method keeps the role. Either way, one of the two parties is now carrying a duty that neither one priced into the contract.
Fix it in the paperwork rather than in an argument later. Every production agreement should state which party is the deployer for the delivered content, who writes the disclosure, and what happens to liability if the disclosure is missing.
What disclosure looks like in a video
The law names the duty and leaves the form open. Accepted approaches include a persistent visual label, an opening disclaimer on the video, and an audible statement. Nothing requires a warning sticker across the frame.
There is a narrower carve-out for work that is evidently artistic, creative, satirical or fictional: the disclosure still exists, but it can be made in a way that does not hamper the display or enjoyment of the work, which is how a film ends up disclosing in its end credits. A product ad is not covered by that carve-out in any reading we would want to defend to a client.
Three formats hold up well for commercial video:
- A line in the end card: "Produced with generative AI." Short, permanent, survives re-uploads because it is burned into the picture.
- A line in the caption or description, for social placements, plus the end card. Captions get stripped when a video is reposted; the end card does not.
- A line in the press kit or product page, for films that get embedded on a site you control.
One detail people miss: content that was generated before 2 August 2026 but published after it still needs the disclosure. Old footage in a new campaign is a new publication.
Canada has no equivalent, which is not the same as no risk
Canada requires nothing comparable today. The 2023 voluntary code of conduct encourages identifying systems that could be mistaken for a human and carries no legal force. The Artificial Intelligence and Data Act died when Parliament was prorogued in January 2025. The recent legislative movement has been election-specific: Quebec's Bill 24 in June 2026 and the federal Bill C-25 that received royal assent on 18 June 2026 both deal with deepfakes in electoral contexts, not with advertising.
So a Montreal or Toronto company that sells only domestically has no labelling duty at all. A Montreal company with a distributor in Germany does, because the obligation follows the market where the content is put in front of people, not the head office.
Other jurisdictions have moved in the same direction on different clocks. South Korea's AI Basic Act took effect on 22 January 2026 with a tiered rule: invisible watermarks for stylized synthetic content, visible labels for realistic content. India's amended IT Rules came into force on 20 February 2026 and require both a visible, persistent label and machine-readable metadata. California's AI Transparency Act became operative on 2 August 2026, the same day as the EU rules, and puts provenance duties on large generative AI providers. California separately requires disclosure when an ad uses a synthetic performer, which we covered in a piece on SB 1050.
Disclosure rules in 2026
When the rules named in this article take effect
| Date (2026) | Jurisdiction | Rule |
|---|---|---|
| 22 Jan | South Korea | AI Basic Act |
| 20 Feb | India | Amended IT Rules |
| 2 Aug | EU | Article 50 applies |
| 2 Aug | California | AI Transparency Act |
| 2 Dec | EU | Marking deadline for systems already on the market |
What to do this quarter
If you run marketing at a company that sells into any of these markets, four things are worth doing before the next campaign ships.
- List every asset in market that was generated or materially edited with AI, with its publication date and its target markets. You cannot disclose what you have not inventoried.
- Name the deployer for each piece of work, in writing, in the contract with whoever produces it.
- Pick your disclosure wording once, put it in the brand guidelines, and stop re-litigating it per campaign.
- Ask your production partner what their outputs carry in metadata today. Providers are being pushed toward machine-readable marking; you want to know whether your files already have it before a platform starts reading it for you.
The part that is easy to miss
Disclosure is being treated as a compliance cost, and it will be, for the companies that do it at the last possible moment in the smallest possible font. It reads differently when a brand does it on purpose. Audiences already expect it: in Hub Entertainment Research's 2025 study, close to 90% of respondents said companies should disclose when AI plays a role in creating content.
The work that survives that disclosure is the work that was good enough not to need the cover. That is a production standard, not a legal one, and it is the one we hold ourselves to. Ours is written down on our Responsible AI page.
Gabriel Brien
Founder of Crimson Spark Agency. AI filmmaker and creative technologist, writing from running this work daily.
More about Gabriel →