The AI Act paperwork, generated
Seven documents the Act expects a deployer to have. Answer the fields, download the result, hand it to your lawyer.
The document pack, as signed PDFs
Every builder on this page, delivered as signed PDFs prepared for legal review: register, disclosures in three languages, literacy policy, vendor questionnaires, FRIA, technical documentation and DPIA. One purchase for the whole organisation.
Marking sampler (Article 50(2))
Samples up to five of a page's images for C2PA provenance metadata, the machine-readable marking AI-generated content owes from 2 December 2026. Informational: bytes cannot prove which images are AI-generated.
What the seven generators write from your answers, and what the image sampler looks for.
Each of the seven buttons selects a fixed template. Pressing Generate sends the fields to our server, which inserts them into that template and returns the document as HTML and as plain text; no language model is involved. If a required field is empty, all missing fields are reported at once and nothing is built. The register assigns every system a risk class from the use you pick in the list: limited for a chatbot or generated content, high for CV screening, credit scoring, biometric identification and emotion recognition, prohibited for social scoring, minimal for internal productivity tools.
What comes back differs by document. The register and the disclosure texts are finished wording: a dated table of your systems, and one short disclosure text per chosen language for chat, generated content or voice. The literacy policy is a short policy with your company name and the roles you listed. The vendor questionnaire is ten fixed questions. The FRIA, the technical documentation and the DPIA are outlines: seven to nine headed sections each, saying what the part must contain, with only your names and the few fields from the form filled in. Apart from the disclosure texts themselves, every document is produced in English.
The risk class is a lookup on eight listed uses, not an assessment of your system, and no generator decides whether a duty applies to you. The marking sampler reads the page HTML, takes the first five img addresses, downloads the first 128 KB of each and searches it for C2PA and digital source type markers. It reports presence only: it does not validate a manifest and cannot tell which images were generated. CSS backgrounds and images added by script are not seen. An email address entered with a generator is stored with a one-line note and passed to our staff. Have a lawyer review each document before use.
- Who is a deployer under the EU AI Act?
- A deployer is a person or organisation that uses an AI system under its own authority in the course of a professional activity. A company that puts a vendor's chatbot on its website, or uses a screening tool in hiring, is a deployer. The provider is the party that develops the system or places it on the market under its own name, and carries the heavier documentation duties.
- What does Article 50 of the AI Act require?
- Article 50 is the transparency article of the Act. People must be told when they are interacting with an AI system unless that is obvious from the context, synthetic audio, image, video and text output has to be marked in a machine-readable format as artificially generated, and deepfakes have to be disclosed as such. Depending on the case, the duty falls on the provider or on the deployer of the system.
- What is the difference between a FRIA and a DPIA?
- A DPIA is required by Article 35 of the GDPR when processing personal data is likely to result in a high risk to individuals, and it looks at the data: lawful basis, necessity, minimisation and safeguards. A FRIA is required by Article 27 of the AI Act from certain deployers of high-risk AI systems, and it looks at the wider effect on fundamental rights such as non-discrimination. One system can need both.
