Quick answer: the EU AI Act's transparency rules (Article 50) apply from 2 August 2026. They mainly bind the companies that make AI tools, which must mark AI-generated images in a machine-readable way, and businesses and professionals who publish deepfakes or AI text on matters of public interest, who must disclose it. Using AI purely for personal, non-professional purposes falls outside the Act.

This is a plain-English summary for creators, not legal advice. If AI images are part of your business, check the official guidance or ask a lawyer.

What Article 50 requires

Article 50 sets out four transparency duties. Two matter for images:

1. AI tool makers must mark their output (Article 50(2)). Providers of AI systems that generate images, video, audio or text must make sure the output is marked in a machine-readable format and detectable as AI-generated. In practice that means techniques like metadata (for example Content Credentials), watermarks or fingerprints. The duty is on the provider: OpenAI, Google, Adobe, Midjourney and so on, not the person typing the prompt.

2. Deployers must disclose deepfakes (Article 50(4)). Whoever uses an AI system to create or change an image, audio or video that looks like real people, places or events and could be mistaken for real (a "deepfake") must disclose that it was AI-generated or manipulated. For clearly artistic, satirical or fictional work, the duty is lighter: disclose in a way that doesn't spoil the work.

Who counts as a "deployer"

The Act defines a deployer as a person or organisation using an AI system under its authority, except for personal, non-professional activity (Article 3(4)). The European Commission's guidance gives the example directly: a person who makes a deepfake in a personal capacity and shares it on social media is outside the Act. But if you do it as part of a business, trade, freelance work or regular paid activity, you count as a deployer and the disclosure duty applies.

So:

  • Hobbyist sharing AI art for fun: outside the Act (platform rules still apply).
  • Freelancer, agency, shop or brand using AI images commercially: a deployer, and realistic AI images of real-looking people or events need disclosure.

When it applies

  • 2 August 2026: Article 50 applies.
  • 2 December 2026: under the AI Omnibus agreement reached in May 2026, generative AI systems already on the market before August get until this date to meet the machine-readable marking duty in Article 50(2).

The Commission has published guidelines and a Code of Practice on marking and labelling AI-generated content, which fill in the practical detail.

Penalties

Under Article 99, breaking the Article 50 transparency duties can bring fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher, with lower caps for small and medium-sized businesses. The Act asks authorities to keep fines proportionate. Enforcement is by national market-surveillance authorities in each EU country.

Does removing metadata break the law?

The marking duty sits with the AI provider, which must make its output detectable. For you, the question is disclosure: if you're a deployer publishing a realistic deepfake, you must tell people it's AI-made, and visible disclosure (a caption, a label, the platform's own AI tag) meets that whatever the file's metadata says.

Where you're a deployer publishing realistic AI content of real-looking people or events, don't rely on the file to do your disclosing, and don't strip metadata to hide that the content is AI-made. For the wider picture, see is it legal to remove AI metadata from images.

Removing metadata remains a normal privacy step for many files: taking GPS location, camera serial numbers or client names out of photos, for example. The AI Act doesn't change that.

Practical checklist for creators working with EU audiences

  • Using AI images in your business? Treat yourself as a deployer.
  • Realistic AI image of a real-looking person, place or event? Disclose it visibly, and use the platform's AI label.
  • Clearly artistic or fictional? A light-touch disclosure is enough, in a way that doesn't spoil the work.
  • Personal, non-commercial posting? Outside the Act, but Instagram, TikTok and YouTube have their own AI labelling rules.
  • Selling a product that generates images? You're a provider, and Article 50(2) marking applies to you.

Frequently asked questions

When does the EU AI Act's AI labelling rule apply?

Article 50 applies from 2 August 2026. AI systems already on the market before then have until 2 December 2026 for the machine-readable marking duty.

Do I have to label AI images I post on social media?

If you post for personal, non-professional reasons, the AI Act doesn't apply to you, though the platform's own rules may. If you post as part of a business or paid work, realistic AI images of real-looking people or events need disclosure.

What are the fines under the EU AI Act for AI content?

Up to €15 million or 3% of worldwide annual turnover for breaching the Article 50 transparency duties, with lower caps for small businesses.

Does the EU AI Act require metadata in AI images?

It requires AI providers to mark output in a machine-readable, detectable way, which can mean metadata, watermarks or both. The disclosure duty for deployers can be met visibly.

Does the AI Act apply outside the EU?

It applies to AI systems and output placed on the EU market or used in the EU, so providers and professional deployers outside the EU can be covered when they reach EU audiences.