December 2 deadline: machine-readable marks on AI video already on the market
By Alan Magliocca · November 13, 2026
Quick answer: Under Article 50(2) of the EU AI Act, providers of generative AI systems placed on the market before August 2, 2026 have until December 2, 2026 to mark their audio, image and video output in a machine-readable way. If you make event video with AI tools, ask each vendor how it will meet that date, keep the marks intact, and keep your visible disclosure too.
- Article 50(2) requires providers of generative AI systems to mark audio, image, video and text output so it can be detected as AI-generated.
- Systems placed on the EU market before August 2, 2026 have until December 2, 2026; the grace period covers only this marking and detection duty.
- Marks must be effective, reliable, robust and interoperable, and the voluntary Code of Practice expects more than one layer where necessary, such as metadata and watermarks.
- Webinar sellers are usually deployers, not providers, so the deadline mostly lands on their AI video vendors.
- Machine marks don't replace a visible disclosure: deployers can't rely on them alone.
What happened
Article 50 of the EU AI Act has applied since August 2, 2026. One part of it carries a later date. Under Article 50(2), providers of generative AI systems that produce audio, image, video or text must mark the output in a machine-readable format, so it can be detected as AI-generated or manipulated. The European Commission's FAQ on Article 50, last updated July 24, 2026, says the marks must be "effective, reliable, robust and interoperable."
The later date is a grace period. Providers whose systems were placed on the market before August 2, 2026 must meet the marking and detection duty only from December 2, 2026. The Commission says the grace period covers only that duty. Systems placed on the market on or after August 2 were covered from day one.
Some output is excluded, including short strings of numbers, symbols or letters, source code, purely machine-to-machine output, and standard editing assistance.
How to mark is set out in a voluntary Code of Practice, published in final form on June 10, 2026. Jones Day reported that the Code accepts no single technique is enough on its own, so it uses a layered approach with at least two layers where necessary, such as metadata and watermarks. Providers are also expected to offer ways for deployers, authorities and researchers to detect whether content is AI-generated. The Commission says companies that don't sign the Code must show compliance by other means and may face more requests for information.
What it means for people selling with webinars
Most webinar sellers are deployers, not providers. You use an avatar tool, a voice tool or a video generator under your own account. The Commission defines a provider as the company that develops an AI system, or has it developed, and puts it on the market under its own name. So the December 2 date mostly lands on your vendors, not on you.
It still changes your work in three ways:
- Your AI video becomes detectable. From December 2, 2026, output from tools already on the market should carry marks that detection tools can read. Assume anyone checking can tell your presenter clips are AI.
- Your visible disclosure still matters. The Commission says deployers can't rely only on the provider's machine-readable marks. If your AI video counts as a deepfake under the Commission's tests, EU viewers need a clear disclosure at first exposure. That side of the rule is covered in EU attendees must see the AI disclosure first.
- Your editing workflow could matter. If you cut clips, re-export or re-encode vendor video, check with the vendor whether its marks survive those steps. California has also moved to bar removing watermarks, covered in Don't strip AI watermarks from event video in California.
You could become a provider yourself in some cases, for example if you build an AI video tool and offer it to others under your own brand. That's a question for a lawyer, and this is a description of the rule, not legal advice.
What should you ask your AI video vendors before December 2?
Send every vendor that generates audio or video for your events the same six questions:
- Was your system on the EU market before August 2, 2026? If so, will your output carry machine-readable marks by December 2, 2026?
- Which marking layers do you use, for example metadata, a watermark, or both?
- Do you offer a way to check whether a file was generated by your system?
- Have you signed the Code of Practice on marking and labeling AI-generated content? If not, how do you show compliance?
- Do the marks survive the export formats we use, and the edits we make, such as trimming or adding captions?
- Are live, real-time avatar sessions marked the same way as rendered video files?
Keep the answers in one document with the date you received them. If a vendor can't answer, that's useful to know before you build an evergreen event on its output. For a broader plan on switching suppliers, see Video models keep changing: a vendor plan for webinar hosts.
A worked example: an evergreen webinar made in spring
Say you recorded an AI-presenter version of your sales webinar in April 2026 and run it every day as an evergreen event.
- The existing recording. The Commission says content generated before August 2, 2026 doesn't need retroactive labels, though it encourages labeling where possible. Adding a visible AI label now is the low-risk choice.
- New clips rendered after December 2, 2026. If your vendor's system was on the market before August 2, these should carry its machine-readable marks.
- Your disclosure. Machine marks or not, EU viewers of a deepfake need a clear disclosure at first exposure, so the label belongs on the first frame and in the opening line.
What to do this week
- List every AI tool that produced audio or video in your current events, replays, ads and clips.
- Send each vendor the six questions above, and set yourself a reply deadline of November 30, 2026.
- Check one recent AI clip after your normal editing and export steps, and ask the vendor whether its marks are still in the file.
- Add a visible AI label and a spoken disclosure to every AI-presented event that EU residents can attend.
- If you build or resell an AI tool under your own name, book a call with a lawyer about whether you count as a provider.
How BrightStage AI helps
This deadline sits with the tools that generate AI video. BrightStage AI doesn't advertise a machine-readable marking feature, so put the vendor questions above to every tool you use for AI video, including the AI presenters in its video creator. Where BrightStage does help is the visible side: you write the registration and thank-you pages, reminders and the video creator's script, which is where your disclosure line goes. For why marks make disclosing first the safer choice, AIVX Labs covers how anyone can now check for AI watermarks.
Sources
Frequently asked questions
Who has to meet the December 2, 2026 deadline?
Providers of generative AI systems that were placed on the EU market before August 2, 2026. The grace period covers only the machine-readable marking and detection duty in Article 50(2).
Do I have to relabel AI video I made before August 2, 2026?
The Commission says content generated before August 2, 2026 doesn't need retroactive labeling, but it encourages labeling it where possible.
If my vendor marks the video, do I still need a visible disclosure?
Yes, for deepfakes shown to people in the EU. The Commission says deployers can't rely only on the provider's machine-readable marks.
Related reading
Written by Alan Magliocca, founder of BrightStage AI and AIVX Labs (Podavinci LLC). Written with AI assistance and checked against the linked sources. Follow along: LinkedIn · YouTube · TikTok · X · Facebook · Instagram · Skool community
