Do the EU AI labeling rules apply outside Europe?

They bind AI providers and deployers whose systems reach people in the EU, which covers every major AI company — and like GDPR, the labeling requirement is likely to become the global default.

Unlabeled AI — The AI Label Law, Explained

Q: Does a missing label prove content is human-made?

No. Older content, non-compliant tools, and stripped metadata all produce unlabeled files. Presence of a valid Content Credential is strong evidence of origin; absence proves nothing either way.

What the law actually says

Article 50 of the EU AI Act is short, but it changes three everyday things. In plain English:

RULE 1

Chatbots must confess

Any AI system that interacts with people must make clear you're talking to a machine — unless it's already obvious from context. The era of "was that a human agent?" support chats is legally over in the EU.

RULE 2

Synthetic media must carry a mark

Providers of AI that generates images, audio, video or text must mark outputs in a machine-readable way — embedded provenance data or watermarks that software can detect, even when your eyes can't.

RULE 3

Deepfakes must be disclosed

If content shows real people, places or events doing things that didn't happen, the person publishing it must disclose that it's artificially generated or manipulated — visibly, not buried in metadata.

The dates that matter

The AI Act rolls out in stages — and the 2025–26 "Digital Omnibus" pushed some deadlines back. These are the ones to watch:

June 10, 2026

The official AI icons arrive

The European Commission's AI Office published its Code of Practice on Transparency of AI-Generated Content, with a standardized set of icons for marking AI-made and AI-edited media. "Labeled" now has an official look.

August 2, 2026

Article 50 applies

The transparency obligations become applicable: AI must disclose itself, synthetic media must be machine-readably marked, deepfakes must be disclosed.

December 2, 2026

The grace period ends

AI systems already on the market before August 2, 2026 got four extra months to comply with the watermarking obligation. After this date, "we launched before the rules" stops working.

Dec 2, 2027 / Aug 2, 2028

High-risk rules follow

Obligations for stand-alone high-risk AI systems (Annex III) were deferred to December 2, 2027, and for AI embedded in regulated products (Annex I) to August 2, 2028.

How AI content gets labeled

Four labeling systems are converging — and every major platform is already using at least one of them:

C2PA Content Credentials

Cryptographically signed provenance data embedded in the file: which tool made it, when, and its edit history. Backed by a coalition of thousands of members including Google, Microsoft, Adobe, Meta, OpenAI and major news agencies. The BBC, AP, Reuters, AFP and The New York Times now publish photos with embedded credentials.

Google SynthID

An invisible watermark woven into the pixels or audio itself — reported to survive screenshots, resizing, recompression and color grading. Google says it has watermarked over 100 billion images and videos, and previewed a public Content Detection API in 2026.

The EU's official icons

Since June 2026 there's a standardized visual mark for AI-made and AI-edited media, published alongside the Commission's transparency Code of Practice — designed so a label means the same thing everywhere in the EU.

Platform labels

TikTok runs a tiered AI-label system with penalties for unlabeled synthetic content and has labeled over a billion videos with provenance data. YouTube, Meta and others require disclosure of realistic synthetic media — proactively labeled content is treated far more gently than content caught unlabeled.

How to check any image or video yourself

You don't need to be a forensics lab. Work down this list — it's ordered from free-and-definitive to judgment-call:

Inspect its Content Credentials Free

Upload the file to the C2PA coalition's official Content Credentials Verify tool. If credentials are present, you'll see exactly which AI tool made it and what's been edited. Absence of credentials doesn't prove it's human-made — but presence is close to proof of origin.

Look for the platform's own label Free

On TikTok, YouTube, Instagram and Facebook, check for an "AI-generated" or "synthetic media" badge near the caption. Missing badge on obviously synthetic content is itself a signal — proactive labels are now the norm for legitimate creators.

Run a detector for unmarked content Tools

When there's no embedded label, detection tools estimate whether content is AI-generated from the content itself. Treat results as probability, not verdict — current consumer-grade detectors run roughly 70–90% true-positive rates with real false-positive rates on genuine photos.

Originality.ai Affiliate

The detector we use for unmarked text and articles — per-scan credits, no subscription required to start.

Try it →

Apply the classic tells Free

Reverse-image search for the original. Check hands, text-in-image, reflections, and physics across video frames. Ask: who posted this first, and does that account exist for any other reason? Provenance of the account often settles what analysis of the pixels can't.

Enforcement is already real

"Unlabeled AI" isn't a hypothetical offense — regulators and platforms are already acting on it:

543,000+pieces of unlabeled AI content ordered removed in China's first mass enforcement action (2026)

13,000+social media accounts removed in the same action across Weibo, Douyin, Bilibili and WeChat

1.3B+videos labeled with AI provenance data by TikTok alone

€15M / 3%maximum EU AI Act fines for transparency violations — €15 million or 3% of global turnover, whichever is higher

Frequently asked questions

What is "unlabeled AI"?

Unlabeled AI is AI-generated or AI-manipulated content that doesn't carry the disclosure or machine-readable marking the law or a platform requires — a chatbot that doesn't say it's a bot, a synthetic image with no embedded provenance mark, or a deepfake posted without disclosure.

Is it illegal to post AI content without a label?

In the EU, from August 2, 2026, providers of generative AI must mark outputs machine-readably, and anyone deploying a deepfake of real people or events must disclose it — with fines for transparency violations of up to €15 million or 3% of worldwide turnover. In China, platform rules already require AI-content labels, and mass takedowns have happened. In the US there's no general federal labeling law yet, but platform policies apply everywhere. This site is journalism, not legal advice — for compliance questions, talk to a lawyer.

Do the EU rules apply outside Europe?

The obligations bind AI providers and deployers whose systems reach people in the EU — which in practice covers every major AI company. Like GDPR before it, the labeling requirement will tend to become the global default, because maintaining separate unlabeled pipelines for non-EU users costs more than labeling everything.

Can't people just strip the watermark?

Visible labels and metadata can be stripped; that's why the law requires machine-readable marking, and why systems like SynthID embed the mark in the pixels or audio itself, where it's reported to survive screenshots, recompression and edits. Stripping a required mark doesn't make content legal — it makes it unlabeled AI with intent.

My AI content was made before August 2026 — am I affected?

AI systems placed on the market before August 2, 2026 got a grace period until December 2, 2026 for the watermarking obligation. After that, the marking rules apply regardless of when the system launched.

Does a missing label prove content is human-made?

No. Absence of a label proves nothing in either direction — older content, non-compliant tools, and stripped metadata all produce unlabeled files. That's exactly why the term "unlabeled AI" matters: it names the gray zone where verification tools and judgment take over. Presence of a valid Content Credential, on the other hand, is strong evidence of origin.

The label era is just starting.

Enforcement begins in earnest on December 2, 2026. We're tracking every takedown, fine, and platform policy change.

Watch: how to spot unlabeled AI →

Disclosure & disclaimer: This site is independent journalism and consumer education, not legal advice. Some links on this page may be affiliate links, meaning we may earn a commission if you purchase through them — at no extra cost to you. Free verification options are always listed first.