AI content labelling policy (Art. 50 AI Act)

This policy sets out when and how CoinPaprika (the Company) labels the content it publishes that was generated or modified by AI, to comply with Article 50 of Regulation (EU) 2024/1689 (the AI Act). These obligations apply from 2 August 2026, and the Digital Omnibus did not postpone them for deployers of AI, which includes the Company. Content generated before that date does not need to be labelled retroactively, with one exception: text generated earlier but published later is covered by this policy.

The policy follows a precautionary principle: when in doubt whether content falls under Article 50, we label it.

The policy applies to everyone who creates, commissions or publishes content on the Company's behalf: employees, contractors, agencies and freelancers. It also covers automations, that is, scheduled jobs that publish content without human involvement.

The policy covers four forms: images, video, audio and text, including when they are combined.

Out of scope:

• internal material the Company does not publish,
• charts, tables and images assembled from market data with a template, without a generative model,
• minor technical editing that does not change the content: spelling and grammar correction, format conversion, cropping, noise reduction.

Partner content published on the Company's channels (e.g. sponsored posts) is treated as our own, as a precaution; the partner declares whether they used AI. For third-party content that we only share, Article 50 places no obligation on the Company, but we do not remove existing labels.

Terms follow the AI Act and the Commission's explanations of Article 50.

Labelling is mandatory in two cases: when we publish a deep fake, and when we publish AI-generated text on matters of public interest whose substance has not been reviewed by a human. In all other cases labelling is voluntary, and the Company recommends it.

Read the tree from the top; the table below gives examples from the Company's channels.

The editorial control exception requires a real review by a person who knows the subject. The review must be the last substantive step: a later substantial change made by AI cancels the exception. The same rules apply to translations, summaries and paraphrases produced by AI. The code of practice expects a documented review procedure; the Company documents the review of every text.

The label can be read without clicking, is visible at the latest at first exposure, and is built into the content itself so that it survives downloading and resharing. Information hidden in the terms of service, in a menu or in a general note about the use of AI does not meet the requirement. Nor does the Company consider a footer or a profile description sufficient.

Common rules:

• The label is in the same language as the content.
• The EU icons come in three variants: basic, “fully AI-generated” and “partially AI-modified”. Using them is voluntary. The Company uses the icon together with a written caption, because the Commission's research shows it is better understood that way.
• We always switch on platform labels (e.g. “AI info” on Instagram and Facebook). The guidelines allow relying on them, but responsibility stays with the Company, so a label in the content itself remains the standard.
• We do not remove machine-readable markings or metadata added by an AI tool. A machine-readable marking does not replace a visible label.
• Accessibility: information about AI goes into the alt text and ARIA labels, and stays on screen long enough to be read.
• In evidently artistic or satirical works, the label may be discreet, e.g. a smaller icon in a corner or a note in the caption.

Ready-made wording in Polish and English is in the “Label wording” section below.

The author of the content is responsible for labelling it correctly, and the approver for clearing it for publication. The same person does not hold both roles for the same piece of content.

Errors and corrections. We fix a missing or incorrect label within 24 hours of finding it. If a platform does not allow editing, we remove the content and republish it with a label. We document every case, together with its cause.

Reports. You can report a missing or incorrect label through our contact form. Reports go to the policy owner.

Review. The policy owner reviews this policy every 6 months and after every change to the Commission's guidelines, the code of practice or KRiBSI's position.

• European Commission: questions and answers on Article 50, guidelines on Article 50 (full text), code of practice, EU icons
• Text of the regulation: Article 50 AI Act
• Poland: Act on Artificial Intelligence Systems (Journal of Laws 2026, item 1003)