AI transparency
You are interacting with AI whenever Chapbook shows an AI/✦ label. Partner replies, draft assistance, editorial passes, generated images, figures, stickers, vector clean-up, PDF/vision import and similar results may be synthetic or inaccurate. Chapbook keeps a human in control of insertion, editing and publication.
How it works
RQAI does not train a general-purpose model on your drafts. You connect Anthropic, OpenAI, Google Gemini or Groq with your own key and choose the active model. A request may include the prompt, conversation history, relevant draft/profile text, an image or source material needed for the selected feature. Requests go directly to the provider. Free image generation, where configured, uses a Cloudflare Worker/Workers AI binding. Private Whisper dictation is local and is not generative AI; browser Live dictation may use a browser-vendor speech service.
Roles under the EU AI Act
The model company is generally the general-purpose AI/model provider. RQAI provides the Chapbook AI interface/system under the Chapbook name. A person or organisation applying output in a professional activity may also be a deployer. This is a current product assessment, not a regulatory certification or a determination for every use case.
Article 50 transparency controls
- Direct interaction: the Partner and AI entry points identify AI from the beginning; replies are labelled and never impersonate a human contributor.
- Human review: output is a suggestion. The user decides whether to insert, modify, discard or publish it.
- Synthetic content: generation surfaces disclose their AI nature. Users must preserve/add disclosures required by law or platform rules.
- Machine-readable marking: provider capabilities and legal exceptions vary. Chapbook does not promise that every model embeds a durable provenance mark in every text/image export. Applicability and provider/export support must be assessed by model and format.
- Accessibility: notices use visible words and are not communicated only by colour.
EU AI Act Article 50 transparency duties apply from 2 August 2026. The limited 2 December 2026 grace period concerns certain pre-existing systems' machine-readable marking, not the obligation to tell people when they are interacting with AI.
Public-interest text, deepfakes and disclosure
If AI-generated/manipulated text is published to inform the public on a matter of public interest, EU law may require clear disclosure unless it received human review/editorial control and a person or organisation holds editorial responsibility. Realistic generated/manipulated image, audio or video may require disclosure as synthetic/deepfake content. Artistic, fictional or satirical context can change how disclosure is displayed but does not automatically excuse deception. The publisher is responsible for the final label.
Limitations and human oversight
- Models can invent facts, quotations, sources, citations, code or events and can reproduce bias.
- Safety filters can fail or over-block; model availability and terms can change.
- Generated interactives/code can contain bugs. Chapbook sanitises and sandboxes previews, but exported work must be tested.
- AI cannot determine copyright ownership, consent, confidentiality, defamation or legal compliance for you.
- Do not use Chapbook as the sole/decisive basis for employment, education, credit, insurance, benefits, policing, migration, justice, healthcare, emergency or other legal/similarly significant decisions.
Privacy, AI literacy and reporting
Minimise personal data, remove identifiers where practical, review the provider's API retention/training and transfer terms, and complete a DPIA where processing is likely to create high risk. Organisations should train users on privacy, security, bias, model limitations, provenance and fact-checking. Report harmful or incorrectly labelled behaviour to support@rqai.co.uk without including keys or unnecessary confidential content.