1. What exactly applies as of today?
Five duties are now in force. Chatbots must identify themselves as AI where that is not obvious. Providers of AI generators must mark their images, videos, audio and text as artificial in a machine-readable way. Anyone spreading deepfakes - AI content that deceptively resembles real people, places or events - must disclose that. AI-generated texts on matters of public interest need a notice, unless a human takes editorial responsibility. And whoever deploys emotion-recognition or biometric-categorisation systems must inform the people affected - though for your feed, the deepfake rule is the one that matters most.
Importantly, the disclosure has to be recognisable at first contact with the content. A hint buried in the fine print does not count.
2. Wasn't all of this postponed?
No - and right now this is the most dangerous misunderstanding out there. The EU Commission's Digital Omnibus in late 2025 mainly pushed back the duties for high-risk systems, to late 2027 and 2028. The labeling obligation for deepfakes and AI content was explicitly carved out and applies as of today. Only one aspect got a grace period: providers whose systems were already on the market before the deadline have until December 2, 2026 for the technical, machine-readable marking.
3. Do I have to label everything I post privately now?
Relax: purely private, non-professional use is exempt. Sharing an AI image in the family chat breaks no rules. The obligation targets those who spread AI content professionally or with reach - companies, media, agencies, influencers, political actors. But be careful: the line is crossed faster than many think. A monetised account with a large audience is not a family chat. Art and satire get a softer form of labeling that must not spoil the work.
4. What do the labels look like?
You will run into three kinds. First, platform labels like "AI-generated" or "Creator labeled as altered" on TikTok, Instagram and YouTube. Second, notices inside the content itself, say an "AI-generated" overlay in a video or a line in the caption. Third - invisible but technically the most important - machine-readable signals inside the file: provenance credentials like C2PA and watermarks like SynthID. How those standards work is covered in our guide to C2PA, SynthID and watermarks.
5. So no label means it's real now?
No, and this is the single most important point in this article. Only content that providers and spreaders choose to label gets labeled. Scammers won't - for them, today only changes the risk of getting caught. Actors outside the EU are hard to reach in practice. The mass of AI clips already in circulation was created before the obligation and stays unmarked. And machine-readable marks often get lost in sharing: screenshots, re-uploads and editing apps strip them.
So the rule for your feed from today on: a present label is a strong signal. A missing label is no signal at all.
6. What are the penalties - and who enforces this?
The AI Act allows fines of up to 15 million euros or 3 percent of global annual revenue. In Germany, the Federal Network Agency becomes the central supervisory and complaints body - citizens will be able to report unlabeled AI content there too. Realistically, big visible cases will be pursued first; the mass of small violations will not change the internet overnight. This article is general information, not legal advice.
7. How do I find out whether something is AI - with or without a label?
Three steps, as always: check the source first (who posted it originally, do reputable outlets report it?), then the visible signs (we show the typical ones in our guides on deepfakes and AI videos), and when in doubt, get a technical analysis. That is exactly what our free AI check is for: submit an image, video or link - we read provenance data like C2PA, look for forensic traces and give you an honest assessment. Including the cases where the honest answer is "inconclusive".