4 rules for your AI content (before you get fined)

0:00 A lot of what you see on YouTube, Instagram, and TikTok right now breaks a new European law. Not next year, today. Why? Because on the 2nd of August, a new part of EU AI Act switched on. It is called Article 50. It is about telling people the truth. If you make a picture, a video, or a voice with AI, you have to mark it. If it could look real, you have to say so. Not in the caption, not in the description, not hidden inside the file, where people actually see it. Get this wrong, and the fine is 15 million euros. Or 3% of everything your company earns in a year. The strange part, almost nobody is talking about this. Here are four rules in plain English.

0:51 The law is called the EU AI Act. It does not arrive all at once, it arrives in steps. On the 2nd of August 2026, one step went live. It is called Article 50. It is about being honest. Not about safety, not about paperwork, about telling people the truth. Here is the part most news got wrong. Everyone expected the big rules for high-risk AI on these dates. Those were moved. They now come on the 2nd of December 2027. So, if someone told you high-risk rules are live today, they are wrong. What is live today is honesty. Rule one.

1:36 If a person talks to your AI, the person must know it is AI. That is it. That is the whole rule. Think about your website chat, your phone support, your booking assistant. If a customer could think they are talking to a human, you have a problem. The message must come at the start, not after 5 minutes, and it must be easy to see. You cannot hide it in a small gray text at the bottom of the page. The law says, "Clear and easy to notice." So, here is your fix. One sentence at the top of the chat. You are chatting with an AI assistant. That sentence costs you nothing. Missing it can cost you millions. Rule two.

2:24 If your AI creates content, that content must be marked. Pictures, video, audio, text. The mark must be machine readable. That means a computer must be able to read it, not only a human eye. A small logo in the corner is not enough on its own. Now, stop. There are two different things here, and people mix them up. One is your system. One is your content. Different rules. First, your system, the tool itself. If your AI system was already running before the 2nd of August, you get extra time to build the marketing in. Your deadline is 2nd of December, 2026. That date is about the tool, not about your old posts.

3:11 Second, your content. What the tool makes. Anything made and published before that December date does not need a label. You do not go back and add marks to old work. You are not fixing 5 years of posts. So, simply, the tool must be ready by December. What you publish after December must carry the mark. Everything before that stays as it is. Rule three. This one surprises people. Some AI reads faces and voices. It guesses how you feel. Are you angry? Are you stressed? Are you happy? Other AI sorts people into groups by how they look. If your company uses this, you must tell people it is used on.

3:58 And you must tell them at the moment it happens, not in a policy nobody reads. Where does this shows up in real life? Call centers that measures customer mood, shops that count who talks in and guesses their age, job interview tools that score a candidate's face. If that is you, you'll need a notice today. Rule four, deep fakes. If you make a video, a photo, or a voice that looks like a real person and it is not real, you must say so. Same for real place or real event. This is not only for bad actors, this is for marketing schemes, too. Use an AI version of your CEO in an ad, say it.

4:46 There is one softer case, art, comedy, and satire. There you still disclose, but in a way that does not ruin the work. And one more, AI-written text about news or public matters. If you publish that, you must say it is AI, unless a real human checked it and takes responsibility for it. So, what happens if you ignore this? 15 million euros or 3% of your global turnover. Whichever number is bigger. If your company earns 1 billion a year, 3% is 30 million. So, you do not pay 15, you pay 30. The percentage is the dangerous part for big companies.

5:36 Now, who pays? This is where people get it wrong. It is not only the company that built the AI. If you only use the AI, you can still be responsible. For chatbots, the builder carries it. For emotional AI and deep fakes, the user carries it. So, we just bought the tool is not the defense. Let us make this simple. Five questions, answer them today. One, does every AI chat we run say it is AI at the start clearly? Two, does everything our AI generates carry a machine-readable mark? Three, do we use anything that reads emotion or sorts people by how they look? Four, do we publish AI images, video, or voice of real people?

6:28 Five, do we publish AI-written text about news about a human editor? If you answered yes to four or five and you have no notice, fix this week. If your AI system is older than August, your deadline is the 2nd of December. This is your date. Put it in your calendar now. Four rules, one deadline, one number to remember. This is not the hard part of the AI Act. The hard part comes in 2027, but this part is live now and it is the easy one to fail. If you want the checklist as a file, the link is in the description. You can find it in my free school community. And tell me in the comments which of four rules is your company breaking.

7:16 Bye.