The EU Artificial Intelligence Act, which kicked in on August 1, 2024, regulates the use of artificial intelligence in Europe. It’s the first comprehensive act governing AI use anywhere in the world.
As the regulatory organization is phasing in its various requirements periodically, certain parts of the EU AI Act will take effect this year.
Why regulate the use of artificial intelligence?
Good question.
Uh, have you read the news?
The internet’s full of stories of AI doing harm to humans. As deepfakes flourish, videos of things that never happened spawn and do real and lasting damage to real people and their reputations.
Scam artists wielding AI tools can replicate voices and put words in your mouth. Let’s say a relative calls you up out of the blue and begs you to send money or else something bad will happen. (Okay, with some relatives, you wouldn’t mind if they suffered a bit, but let’s err on the side of good-heartedness.) So was it real? Most likely, someone prompted an AI bot to ape your relative’s voice, to call you, to exploit your loyalty, and to take your bank account to the cleaners. If you send money, your money’s gone for good.
Then there’s disinformation and propaganda. AI can disseminate false news, fan flames of biases, and incite people to do rash things. If you live in a democratic country, and someone, somewhere, wants to topple the government, or at least topple trust in it, AI’s the likely weapon of choice.
What the new law will cover
On August 2, 2026, all new AI systems and generative tools need to comply with the EU AI Act’s Article 50 transparency requirements. This particular article mandates that EU media companies must disclose and label all segments of AI-generated content the very moment that the segments are released. That’s big. This part of the EU AI Act affects all sorts of companies that deal in and deliver media, including those in the streaming industry.
Four months later, on December 2, 2026, a grace period will end for any AI systems, audio, images, and video that happened to be on the market before August 2, 2026. So then they, too, must conform.
Finally, on February 2, 2027, media providers and platforms are required to have technical solutions up and running to make sure that watermark detection mechanisms are interoperable across different systems.
Which content needs to be labeled?
Deepfakes
This category means audio, image, or video content that has been generated or manipulated to look or sound like real people, objects, or events, and that appears authentic, but is actually not. Chatbots fall under this category.
Public interest text
This class of labelable content includes AI-generated or AI-manipulated text that is published to inform the public on matters of public interest.
So instead of banning AI, the law seeks to inform the public that AI is being used. The EU AI Act shall guide people to use their own discretion when viewing and listening to media that has been labeled as AI-generated or AI-modified.
You’re a media company: now what?
How can the streaming industry hope to hew to the new guidelines? The disclosure and labeling requirements state that reporting synchronicity is mandatory. Companies must disclose and label the content as altered by AI at the same time as the release of the content. That’s a lot to handle, so here are some best practices that you can follow to make sure you're on the straight and narrow when it comes to AI use and its disclosure.
Take stock of current AI use
Media companies and their supporting platforms should take a look at each and every AI application that they use throughout their workflows, from ideation to delivery. Listing and comprehending the nooks and crannies where AI toils away in the whole process of shipping content will help organizations comply with the new rules. It’ll also help specify which transparency obligations apply to which steps in the content delivery pipeline.
Define your AI policy
If you’ve got a plan for approving how AI may be used throughout your company’s content production process, then you’re set up for success. Determine the stages at which AI tools may be used, and where human oversight is necessary. Also pinpoint who’s monitoring AI-assisted content before posting it, and how disclosures should be applied. With a proactive plan, you won’t have to backtrack and do retrofitting. This way, transparency requirements are baked into the process, rather than added, after the fact, as an afterthought or a tack-on.
Better yet, integrate compliance labeling as a technology into your CMS (content management system) so that your disclosure labels are automatically added. Then you won’t have to worry about doing it manually.
Prepare to label
Going forward, deepfakes and other “created-out-of-thin-air” content aren’t illegal. They’re just subject to mandatory disclosure. So inform your audiences what’s synthetic. Just let them know, that’s all.
Reveal AI-generated text
Some people publish text that AI wrote for them without so much as glancing at it to make sure it’s fact-based. Here’s the thing. AI can be a powerful tool to use when iterating, correcting grammar, catching typos, and so on. But paying little to no attention to the text you had AI generate for you before posting it means you’re not exercising genuine editorial control. And that’s okay, as long as you’re willing to catch any flak for mistakes. But not tagging such text as AI-generated won’t fly with the EU AI Act.
Train staff
Time to circle the wagons! Get everyone on the same page about your company’s policies and the legal ramifications of non-compliance. The best understanding within your organization is a shared understanding. That way, everyone knows about their organization pushing out synthetic media, cloned audio, and AI-generated content in general, and the lawful way of labeling it all.
Document, document, document
Keep records that demonstrate how you’re complying with the AI Act’s requirements. So if regulators come a-knocking, you’ve covered all your bases.
Talk to your third-party vendors
Your company’s busy. You probably outsource at least one aspect of content generation or media production. So go over your contracts with your vendors with a fine-toothed comb to make sure they are providing the right information that supports compliance with the EU AI Act. That may mean double-checking that your vendors are providing all the documentation about how AI-generated content is identified and the provenance metadata is embedded within it.
Preserve AI provenance info and metadata
Speaking of provenance metadata, you gotta lock that down. Technical markers that trace origins of content and the alterations that occur during the content’s arc through the internet are available. One such technology along this line is the Coalition for Content Provenance and Authenticity (C2PA) standard, an authority on establishing the credentials of content.
Unified Streaming has worked closely with C2PA to develop the specification and hone it for streaming. This year at IBC, the yearly media tradeshow in Amsterdam, Unified Streaming will demonstrate a new solution called Trusted Media.
Unlike some early C2PA implementations, which have mainly addressed just single, static images, Trusted Media, harnessing the power of C2PA, enables authenticity information to be added dynamically to video during the final stage of content preparation. That’s before the media moves to the CDN. The streaming player or device (such as a smartphone, smart TV, or tablet) then validates the content in real time.
Unified’s also been busy supporting the EBU, or European Broadcasting Union, in its efforts to inform the public about content origins, content credential labeling, and better awareness of AI-generated and AI-modified audio and video.
To see the Trusted Media solution in action, come to IBC 2026 in Amsterdam, September 11–14, booth 5.C68. Or just set up a meeting. We’re eager to talk about new developments with you.
Last thoughts
Sure, the EU AI Act has good intentions, but that doesn’t mean it’s gonna be a barrel of laughs. Compliance sounds serious for a reason.
Like with most things, going in with the mindset of “well begun is half done” may help you succeed and relieve stress at your organization. We’re here to help. Together, maybe we can look forward to a future where the currency of authenticity wins out, and AI’s ability to fool us gets tamped down. (Dare to dream.)