Classified examples · Building AI products · Po polsku
We build a customer chatbot on a model API
Transparency (Article 50)Provider
The product team builds a support chatbot on an external model API (for example OpenAI, Anthropic or Mistral) and launches it under the company's brand.
How we classify it
- The company develops an AI system and puts it into service under its own name: it is the provider of that AI system. The model vendor remains the provider of the general-purpose model.
- Customer support is not high-risk and not prohibited.
- As provider, the company must design the chatbot to tell people they are dealing with AI (Article 50(1)) and make sure generated content is marked in a machine-readable way where Article 50(2) applies: check what the model provider already marks and add what is missing.
- Result: provider with transparency obligations since 2 August 2026; marking for systems already on the market by 2 December 2026.
What applies now
- AI disclosure at the first message, and handover to a person
- Marking of generated content, tested end to end (streaming, exports)
- Limits against prompt injection; least privilege for any tools the bot can call
- Logs and retention agreed with data protection; vendor terms on training and retention
- Register entry, incident process
What would change the answer
- If the chatbot decides on credit, insurance, jobs or essential services, the system becomes high-risk and the company carries the provider obligations for high-risk systems
Evidence to keep
- Transparency checklist
- Test records for disclosure and marking
- Vendor questionnaire
- Incident procedure
In the AI Compliance Kit: 05Transparency Checklist and Disclosure Texts · 03AI System Inventory · 07Vendor Due Diligence Questionnaire · 08AI Incident Response Procedure.
Dates and sources
- Transparency obligations, Article 50: 2 August 2026
- Machine-readable marking of generative AI output: 2 December 2026
Legal basis: AI Act art. 3(3), 50(1), 50(2).
Similar examples
- We fine-tune an open-weight model for our product Depends on the use