August 2, 2026: the date when AI content labeling goes from best practice to legal obligation. On that date, the EU AI Act's Article 50 and California's amended AI Transparency Act (SB 942) both go into effect.
That means conversational AI platforms, on-device generative editing tools, and AI-assisted marketing creative are subject to a mandatory set of AI transparency regulations. Taken together with China and South Korea's existing rules, 2026 is the year AI disclosure stops being optional.
When did the EU AI Act transparency requirements take effect?
On August 2, 2026, Article 50 came into force for existing systems and new launches. It covers:
- Chatbots and virtual assistants: Users must know they're interacting with AI.
- AI-generated content: Outputs must include a disclosure stating that they were created by AI.
- Deepfakes: Artificially generated or manipulated images, audio, and video must be labeled.
- Emotion recognition and biometric categorization: Use must be disclosed.
Under this AI transparency law, AI providers must label certain outputs in a machine-readable format. Professional deployers must also disclose synthetic content to users.
Penalties: Fines up to €15 million or 3% of global annual turnover, whichever is higher.
Note: existing generative AI systems get a short runway on one requirement. Under the May 2026 Digital Omnibus agreement, systems already on the market before August 2, 2026 have until December 2, 2026 to implement machine-readable marking of synthetic outputs. All other Article 50 obligations, including chatbot disclosure and deepfake labeling, apply immediately.
Reach: Any companies with content that reaches EU users, regardless of where the company is headquartered.
Does California's AI Transparency Act (SB 942) now apply to U.S. companies?
Yes. SB 942 was originally set to take effect January 1, 2026, but California delayed it to August 2, 2026, the same date as the EU's deadline.
The requirements roll out in stages:
Now: Any generative AI product with more than 1 million monthly California users must:
- Offer a free tool for the public to confirm whether content came from that system.
- Allow users to add visible AI-generated content labels.
- Include a latent disclosure (such as a machine-readable watermark) in AI-generated images, video, and audio.
January 2027: Social media apps and search engines must detect and surface that watermark data, even for content they didn't generate.
January 2028: Cameras and recorders must let users tag content as authentic at the point of capture.
Penalties: $5,000 per violation, per day
Reach: A system doesn't have to be built in California – only publicly available to 1 million+ California users.
What AI content labeling laws are already in effect in China and South Korea?
Both countries have laws that are currently being enforced.
AI transparency laws in China
Since September 1, 2025, the Cyberspace Administration has required both visible labels and embedded metadata for all AI-generated text, images, audio, video, and virtual scenes (per national standard GB 45438-2025). Major platforms like WeChat, Douyin, and Weibo already comply.
Penalties: Violations can result in content removal, platform warnings, suspension of content publishing capabilities, and escalation under laws such as China's Cybersecurity Law and Personal Information Protection Law. Serious violations can result in penalties reaching tens of millions of RMB.
Reach: Applies to any AI content distributed to Chinese users.
AI transparency laws in South Korea
On January 22, 2026, South Korea's AI Basic Act took effect. Under this AI transparency regulation, generative AI providers must notify users that AI is involved. That means providers either label outputs with a human-readable notice or machine-readable watermark. Deepfakes need a persistent, human-readable notice for the entire clip or image.
Penalties: Fines of up to ₩30 million (about $20,000 USD) per violation, following a grace period that runs through January 2027.
Reach: Applies to any high-impact or generative AI product offered to Korean users.
AI transparency requirements based on what you build
Conversational and character-generation platforms need to handle chatbot-disclosure obligations under both Article 50(1) and Korea's AI Basic Act, including for any images or voice outputs a chatbot might generate mid-conversation.
Device makers with on-device generative editing are the biggest target of the next regulatory wave. As phones start including retouching, background replacement, or "remix" features, every output will require a provenance trail that survives exports, screenshots, and re-compression. California's 2028 capture-device rule extends this obligation to the point of image capture itself.
Enterprise marketing teams face the broadest exposure: AI-assisted creative can cross all four jurisdictions in a single campaign. Vendor contracts, asset pipelines, and localization workflows all need embedded, tamper-resistant watermarking that will survive resizing and re-uploads.
What do all four AI transparency laws have in common?
Visible labels alone are not enough. The EU, California, China, and South Korea AI transparency laws are all moving toward durable, machine-readable provenance. These laws demand that details about origins stay connected to content, no matter where it goes.
Steg.ai helps organizations meet these challenges. Instead of just adding captions or removable metadata, you can embed tamper-resistant watermarks directly into generated content. Compliance travels with the file, helping teams verify origin, preserve transparency, and evolve to meet new AI regulations.