📌 This article is general guidance, not legal advice. Consult your own legal or compliance team to confirm what your specific use case requires.
Introduction
From 2 August 2026, two transparency regimes apply to AI-generated video: the European Union’s AI Act (Article 50) and California’s AI Transparency Act. They work differently, and the difference matters for what — if anything — you need to do:
California’s AI Transparency Act places its obligations on the provider of the AI system — that’s Synthesia. Synthesia marks AI-generated content in a machine-readable way and makes the required disclosure tools available. The Act does not place obligations on customers who generate videos.
The EU AI Act places a machine-readable marking duty on the provider (Synthesia) and a separate visible-disclosure duty on the deployer — the person who publishes AI-generated or manipulated video that could appear to show real people or events.
What Article 50 Requires
Article 50 of the EU AI Act sets transparency obligations for certain AI systems, and these apply from 2 August 2026. For AI video, two things matter most:
Machine-readable marking: the organization that provides the generative AI system marks its synthetic audio, image, and video outputs so they can be detected as AI-generated. Synthesia does this for you.
Visible disclosure: the organization that publishes AI-generated or manipulated video that could appear to show real people or events discloses that the content is artificial. This disclosure needs to be clear and shown to viewers at the latest when they first see the content.
📚 For the full definitions, obligations, and latest guidance, refer to the European Commission's official overview: Transparency obligations under Article 50 of the AI Act.
💡 What counts as a “deepfake”: a realistic AI presenter can fall within the visible-disclosure rules even if the avatar is not a real, identifiable person. If you’re unsure about realistic-looking video shown to EU audiences, treat the visible disclosure as likely needed and confirm with your own counsel.
Provider vs. Deployer (Who's Responsible for What)
The EU AI Act assigns obligations by role, and the two roles carry different duties.
A provider builds an AI system and offers it under its own name. The machine-readable marking duty sits here.
A deployer uses an AI system for a professional purpose. The visible disclosure duty for AI-generated or manipulated video sits here.
For most Synthesia customers, Synthesia is the provider of the generative system, and you are the deployer of the videos you create and publish. In practice, Synthesia handles the machine-readable marking, and you decide when a video needs a visible disclosure for your audience. As the deployer, you remain responsible for your specific use. The two duties are separate, so a visible disclosure may still be needed even where machine-readable marking is present.
📚 For the official definitions of these roles, refer to the European Commission's AI Act service desk.
What Synthesia Does Automatically
Synthesia helped shape the EU's Code of Practice and Guidelines, and was among the first companies to build these provisions into its platform. Synthesia handles the technical marking so you can meet your own obligations more easily. You remain responsible for your specific use case and deployment context.
Provenance signals (C2PA): relevant videos you generate with Synthesia carry C2PA-compliant provenance signals that mark them as AI-generated, with no setup required from you. These signals are machine-readable — embedded in the file and read by detection tools, not seen by viewers. They satisfy the provider-side marking duty (EU Article 50(2)) and California’s machine-readable marking requirement.
Because these signals are invisible to viewers, they do not, on their own, satisfy the EU visible-disclosure duty that may apply to you as a deployer. That’s what the visible label below is for.
How to Add a Visible AI-Generated Label
The visible on-screen label is separate from the C2PA provenance signals: provenance is machine-readable and added automatically, while the visible label is a human-visible notice you switch on when you need it.
Under California’s AI Transparency Act, this visible label is an option Synthesia is required, as the provider, to make available to you but the choice to use it stays with you, and California does not require you to apply it.
There are two ways to turn it on:
Workspace default (admins): label every video created in the workspace automatically.
Individual (any user): label only your own videos, if the workspace default isn't switched on.
To set the label as a default for your workspace
Navigate to Workspace Settings → Collaboration.
Toggle on AI content watermarking.
An AI label will be added to the top corner of all videos generated in your workspace. This also applies to existing videos — if you edit and re-generate them, the label will be added.
To label your own videos (individual user)
If your admin hasn't enabled the workspace default, you can still label all of your own videos:
Navigate to Account Settings → Preferences → AI transparency
Toggle on Automatically add the AI label to my videos
💡 The visible label and the machine-readable provenance signals are separate. Provenance is added for you with no setup, while the visible label is a choice you make for each video or workspace.
What You Should Check With Your Own Counsel
Synthesia handles the technical marking and gives you the tools to disclose AI-generated content, but you remain the deployer of the videos you publish. That means the decision about what your specific use case requires sits with you. Before you rely on this guide, confirm the following with your own legal or compliance team:
Whether your content falls within the scope of EU AI Act Article 50, based on where your audience is and how the video is used.
Whether a visible disclosure is required for your use case, and how prominent it needs to be.
How the rules apply to videos you created before 2 August 2026, or that you edit after that date.
What records you need to keep to show that your content is marked and disclosed.
Any other laws that apply to you. Transparency rules continue to develop across regions.
✍️ This article is general guidance, not legal advice. Synthesia cannot advise on your specific obligations. For anything beyond the general position, refer to the official sources linked in this article and consult your own legal or compliance team.
💬 FAQS
What's changing on August 2, 2026?
What's changing on August 2, 2026?
Two pieces of legislation start to apply on the same day:
Article 50 of the EU AI Act — the transparency rules within the European Union's AI law
California's AI Transparency Act (SB 942)
In broad terms, both require AI-generated content — including video — to be:
Marked in a machine-readable way, so software can recognise that it was created or altered by AI, and
Clearly disclosed to the people who see it, so they know they're looking at AI-generated content.
Similar rules are emerging in other countries (including the UK, Japan, South Korea, and India), so this is part of a wider global shift toward AI transparency.
When exactly do the rules take effect?
When exactly do the rules take effect?
The core obligations apply from August 2, 2026. A few technical requirements have slightly later deadlines — for example, under the EU AI Act, providers of AI systems that were already on the market before 2 August have until December 2, 2026 to fully meet the machine-readable marking requirement.
Do these rules apply to me?
Do these rules apply to me?
It depends on who sees your content and where. In general:
Under the EU AI Act, if you use Synthesia to create AI-generated video that people in the EU will see, you're typically considered a deployer — which can mean you're responsible for disclosing that the content is AI-generated.
Synthesia is the provider of the AI system, so we carry the provider-side obligations, such as embedding machine-readable provenance in the content you generate.
Because your exact obligations depend on your content, your audience, and how you use Synthesia, we'd always recommend confirming the specifics with your own legal team.
What if my audience isn't in the EU or California?
What if my audience isn't in the EU or California?
These rules focus on content seen by people in those regions. If none of your audience is there, they may not apply to you directly. That said, similar rules are appearing elsewhere and audiences can be hard to predict, so many customers choose to apply the same transparency approach to all of their content. Your legal team can advise on what's right for you.
What does Synthesia do automatically?
What does Synthesia do automatically?
Videos you generate on Synthesia carry C2PA-based provenance signals embedded in the file. These signals:
mark the video as AI-generated, and
record a tamper-evident chain of custody back to the source.
This happens automatically when you generate your video — there's nothing you need to switch on.
What is C2PA / "provenance"?
What is C2PA / "provenance"?
C2PA (the Coalition for Content Provenance and Authenticity) is a widely adopted, cross-industry standard for attaching secure, machine-readable information about how a piece of content was made. Think of it as a verifiable label that travels with the file. Synthesia was among the first companies to join C2PA, and it's the same standard the new EU rules are designed to work with.
Do I need to add a visible "AI-generated" label?
Do I need to add a visible "AI-generated" label?
For content that will be seen by people in the EU or California where a visible disclosure is expected, you can turn on a clear on-screen "AI-generated" label — similar to burning in captions.
Whether a particular video needs a visible label depends on the content and where it will be shown, so it's worth confirming your specific cases with your legal team.
When should I turn the visible label on?
When should I turn the visible label on?
As a general guide, consider adding the visible label when a video will be seen by people in the EU or California.
How do I add the label?
How do I add the label?
You can toggle it on per video, and administrators can manage the controls at the workspace level.
What about videos I created before August 2, 2026?
What about videos I created before August 2, 2026?
There's no requirement to go back and relabel content that was generated before the rules took effect. Under the EU's guidelines, the cut-off is based on when a video was generated or altered, so videos made before August 2, 2026 don't need retroactive marking or labelling. You only need to consider the new rules for content generated or altered from that date onward.
Will labelling make my videos look less professional?
Will labelling make my videos look less professional?
The machine-readable provenance is invisible to viewers — it lives in the file, not on the screen. The visible label, where you choose to use it, is designed to be clear but unobtrusive, and you decide which videos need it.
What are the penalties for not complying?
What are the penalties for not complying?
Under the EU AI Act, non-compliance can carry fines of up to €15 million or 3% of global annual turnover, whichever is higher. In California, the fines can scale to $5000 per day, per violation. These obligations sit with the organisation deploying the content — which is exactly why we've built provenance in by default and made disclosure easy, to take as much of that weight off you as we can.
Is Synthesia ready for these rules?
Is Synthesia ready for these rules?
Yes. This is something we've been preparing for well ahead of the deadline. Synthesia was among the first companies to join C2PA, helped draft Partnership on AI's Responsible Practices for Synthetic Media, and contributed across all three rounds of the EU's Code of Practice on transparency of AI-generated content. We're also independently certified to ISO 42001 (AI management), SOC 2 Type II, and ISO 27001. And we're applying the same transparency standard to our own AI content across our website, help center, and social channels.