Europe’s AI labeling and transparency rules are now in effect
The Verge Jess Weatherbed ● Covered by 75 sources
The EU's AI Act transparency rules kicked in August 2nd. Chatbots and deepfakes now legally need to say what they are.
Based on reporting by The Verge, Jess Weatherbed — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Brussels just flipped a switch that's going to change how a lot of AI content looks online, even if most people won't notice at first. As of August 2nd, companies operating in the EU have to make it obvious when someone's talking to a machine instead of a person, and they have to slap detectable markers on anything AI-generated or altered — audio, video, images, text, all of it.
The rules split responsibility two ways. Providers, the companies actually building these AI systems, have to bake in machine-readable marks so synthetic content can be flagged as artificial. Deployers, the platforms and services putting that AI to use, have to label deepfake-style content designed to pass as real. Some companies, like Meta and SpaceXAI, get hit with both sets of obligations because they play both roles.
The European Commission's reasoning is pretty blunt: generative AI has gotten good enough that people can't reliably tell what's real anymore, and that erodes trust and opens the door to misinformation. So the Commission built a standard set of disclosure icons companies can use, loosely echoing labels TikTok, Instagram, and Facebook already rolled out on their own. Using the EU's specific icons is optional. Actually complying with the underlying labeling requirement is not, and the Commission was pointed about making that distinction clear.
Money is what will make this stick, or not. Noncompliance risks fines up to €15 million, or 3 percent of a company's global annual turnover, whichever hits harder. New AI systems have to comply right away. Anything that was already live before August 2nd gets a four-month cushion, running until December 2nd, before the same penalties apply.
My take — AI-written commentary, not fact-checked reporting
Giving companies a four-month grace period for existing systems is the sensible part of this — nobody wants a scramble that breaks half the internet's chatbots overnight. But the real test is whether a 3 percent global turnover fine actually scares anyone bigger than a mid-size startup, because for the largest AI players that number can look like a rounding error rather than a deterrent. Labeling AI content is the easy, obvious move; making the label mean something when it costs money to ignore it is the hard part nobody's solved yet.
Read more about this at: The Verge
