EU AI Act Takes Effect: AI Content Labels Now Required

The EU AI Act’s transparency rules go live on August 2, 2026. From that date, every AI system that generates video, audio, images, or text in the EU must mark its output in a machine-readable format so the content can be detected as AI-generated. California’s AI Transparency Act (SB 942) kicks in on the same day with parallel requirements. Synthesia signed the EU’s accompanying Code of Practice today, joining Google and other major AI providers who committed to the framework in the past week.

The Two Laws Taking Effect on the Same Day

Two laws take effect on August 2, 2026, and together they cover most of the world’s AI-generated content. The EU AI Act’s Article 50 applies to any AI system operating in the EU, while California’s SB 942 targets tools with more than one million monthly users. Both require machine-readable provenance marks and, in certain cases, visible labels.

EU AI Act Article 50 splits responsibilities between providers (the companies building AI tools) and deployers (the businesses and creators using them). Providers must embed machine-readable marks in AI-generated content. Deployers must visibly label deepfakes and AI-generated text published to inform the public on matters of public interest. The regulation covers audio, image, video, and text.

California SB 942 applies to any AI system with more than one million monthly users accessible in California. It requires three things: a free public AI detection tool, latent (hidden) machine-readable provenance embedded in every AI-generated image, video, and audio file, and the option for users to add visible manifest disclosures.

Who Signed the EU Code of Practice

The Code of Practice on Transparency of AI-Generated Content is a voluntary framework drawn up by independent experts and facilitated by the EU’s AI Office. It was published on June 10, 2026 and assessed as adequate by both the European Commission (July 8) and the AI Board (July 9). Companies that sign it get a recognized path for demonstrating compliance across all 27 EU member states. Those that do not sign can still comply, but must prove independently that their own systems meet the requirements.

Google signed on July 24, 2026, alongside an expansion of its SynthID watermarking technology to five additional AI partners: Apple, ElevenLabs, Kakao, NVIDIA, and OpenAI. Synthesia announced its signing today (July 29), noting that it helped shape the Code during the working groups that drafted it.

The initial signatory deadline was July 27, 2026 for inclusion on the first published list. The AI Office expects to publish that list on July 31. Companies can continue to sign on an ongoing basis after the initial list.

How C2PA and SynthID Work Together

Most major providers are building around two complementary layers: C2PA cryptographic metadata and invisible watermarks. C2PA records who created a file and with what tool; SynthID embeds invisible signals that survive screenshots and compression. Neither works perfectly alone, but combined they satisfy the Code of Practice’s multilayer approach.

C2PA (Coalition for Content Provenance and Authenticity) attaches cryptographically signed metadata to a file, recording who created it, which tool was used, when it was generated, and whether it was edited. Adobe, Microsoft, the BBC, and over 100 other organizations developed it through the Linux Foundation. The catch: a screenshot or platform re-encode strips the metadata entirely.

SynthID, developed by Google DeepMind, embeds invisible signals directly into the content itself. These signals survive compression, cropping, screenshots, and re-recording. SynthID has already watermarked over 100 billion images. But on its own, it tells you almost nothing about the content’s origin.

The two layers compensate for each other. If a platform strips the C2PA metadata (Instagram does this today), SynthID can still flag the file as AI-generated. If the C2PA stays intact, you get the full provenance trail: who made it, when, with what tool.

What This Means for AI Video Creators

If you use AI video generators or other tools to create or manipulate video, audio, or images, your workflow changes on August 2. How much depends on whether the law classifies you as a provider or a deployer. Most individual creators fall into the deployer category, which means the visible labeling obligation lands on you.

Providers (companies building the AI tools) handle the technical side. They need to embed machine-readable marks in every output. Most major platforms already do this through C2PA or SynthID, so if you’re using Synthesia, Runway, or similar tools, the marking happens before you ever touch the file.

Deployers (businesses and creators using these tools) handle the disclosure side. If you publish a deepfake, you need to visibly label it as AI-generated at the moment of first exposure. If you publish AI-generated text to inform the public on a matter of public interest, it needs a label unless a human reviewed it and took editorial responsibility.

Three exemptions apply to deployers:

1. Editorial review exception: AI-generated text that has undergone genuine human review and editorial control does not need disclosure
2. Artistic/satirical exception: Deepfakes in clearly artistic, satirical, or fictional contexts require a lighter label that must not obstruct the work’s enjoyment
3. Technical assistance exception: AI systems performing purely assistive functions (spell-checking, standard editing) are exempt from marking requirements

How Synthesia Built Compliance In

Synthesia took a different approach from companies racing to meet the August 2 deadline. The company participated in the working groups that drafted the Code and built compliance into the platform alongside its ISO 42001 certification and SOC 2 Type II audit.

Videos generated in Synthesia carry C2PA provenance signals by default. Users can view Content Credentials from a menu in Synthesia’s video player, and signed downloads are coming later this year so the credentials travel with the exported file.

For visible labeling, Synthesia added a workspace-level toggle that applies the EU’s reference AI label to every scene. Individual users can also enable labels through their preferences if the workspace toggle is off. The company is working on letting customers change the label position or use their own brand fonts and colors through Brand Kits.

Compliance Timeline

The August 2 deadline is not the whole story. Multiple deadlines stack over the next 18 months, and each one extends the scope of who must comply and what counts as covered content. Here’s the full timeline.

Date	What Happens
August 2, 2026	Article 50 takes effect. Chatbot AI disclosure, deepfake labeling, and public-interest text labeling apply immediately. SB 942 takes effect in California.
July 31, 2026	EU AI Office publishes initial list of Code of Practice signatories.
December 2, 2026	Grace period ends for machine-readable marking of AI systems placed on the market before August 2.
January 1, 2027	California AB 853 extends obligations to large online platforms.
January 1, 2028	California extends obligations to capture device manufacturers.

Content you generated before August 2, 2026 does not need retroactive labeling, though the European Commission encourages it.

Beyond the EU: Where AI Content Rules Are Heading

The EU and California went first, but they won’t be alone for long. Both landed on the same enforcement date and the same basic split: providers mark content at the source, deployers label it for the audience. Other jurisdictions are watching how enforcement plays out before drafting their own rules, which means what happens after August 2 will shape global AI ethics and content regulation for years.

Google flagged a practical problem in its signing announcement: overlapping labels from different regulatory regimes could confuse viewers instead of informing them. If content passes through three AI tools and each adds its own label, you end up with noise.

The EU is addressing this with three standardized icon designs that deployers can use to label AI content. The icons are localized by language (AI in English, KI in German, IA in French) and designed for different media types: persistent on-screen labels for video, visible labels for images, and audible disclaimers for audio.

FAQ

Sources

1. Synthesia - Among the First AI Video Companies to Sign the EU’s Article 50 Code of Practice (July 29, 2026)
2. European Commission - Code of Practice on Transparency of AI-Generated Content
3. European Commission - Transparency Obligations Under Article 50 FAQ
4. Google - Signing the EU AI Act Transparency Code of Practice (July 24, 2026)
5. California Legislative Information - SB 942 AI Transparency Act
6. European Commission - Guidelines on Transparency Obligations
7. C2PA - Coalition for Content Provenance and Authenticity