Skip to content

AI and security

Consumer chatbots and AI workspaces: what is the difference?

This article was created with AI.

Last updated 11 min read

A guide by sicherAI. We offer an AI service for Heilpraktiker practices. How we write these guides

Deutsche Fassung

In short

A consumer chatbot and an AI workspace differ in the contract, which sets out who decides how your inputs are used: in a personal account, the provider sets its own purposes, even on a paid plan. In an AI workspace on a business plan with a data processing agreement, the provider commits to act as a processor and to handle personal data on the practice’s documented instructions, though the agreement can reserve some purposes for the provider itself. Whether training is switched off by default depends on the plan. Without a data processing agreement, personal data about patients does not belong in a provider’s chatbot.

Contents

What is an AI workspace?

An AI workspace is a business plan in which a practice manages its team’s accounts under the provider’s business terms. Its counterpart is the consumer chatbot: a chat app that one person uses with a personal account under the provider’s terms for individuals. A data processing agreement, in German Auftragsverarbeitungsvertrag or AVV, is the contract that Article 28(3) of the General Data Protection Regulation (GDPR) requires when a provider processes personal data on someone else’s behalf.

The GDPR calls the party that “determines the purposes and means of the processing of personal data” the controller (Article 4(7) GDPR). The processor acts on the controller’s behalf (Article 4(8) GDPR) and “only on documented instructions from the controller” (Article 28(3)(a) GDPR).

Every practice of Heilpraktiker in Germany that drafts emails or letters with AI help faces this choice. Heilpraktiker are non-medical practitioners with a state permit under the German Heilpraktikergesetz to practise medicine without being licensed as doctors. Some hold a permit restricted to psychotherapy.

Which kinds of AI service can a practice choose from?

A practice can choose from three kinds of AI service, each with its own contract type: a personal account in a chat app, a business plan with a data processing agreement, and a system the practice runs itself or through an IT firm. They differ in who decides how inputs are used.

Personal accountBusiness planOwn or local system
ContractTerms for individualsBusiness terms with a data processing agreementSoftware licence; data processing agreement with any IT firm running it
Role of the providerController that sets its own purposesProcessor under the contract, on the practice’s documented instructions; the agreement can reserve some purposes for the providerNo outside AI provider for the model
Use of inputs for trainingPossible unless the user opts out; some terms still allow use of feedbackSet by the business terms: often excluded by default, sometimes only after an opt-out; exceptions such as feedbackDecided by the practice
Who manages the accountsEach person for themselvesAn administrator of the practiceThe practice

For all three kinds, the Federal Office for Information Security (BSI) notes that using generative AI models “also introduces novel IT security risks and can amplify known cybersecurity threats”.

Does one provider offer both personal and business plans?

Yes, several providers offer both personal accounts and business plans with a data processing agreement; whether those plans contractually exclude training varies. The list is alphabetical and ranks no provider.

  • Anthropic: its Consumer Terms (effective 8 October 2025) cover services “for individuals”, such as Claude.ai and Claude Pro. Its Commercial Terms (effective 17 June 2025) incorporate a Data Processing Addendum under which “Customer is the controller and Anthropic is Customer’s processor”, and state that Anthropic “may not train models on Customer Content from Services”. Its Privacy Center (as of 6 November 2026) names explicitly reported feedback as an exception.
  • Mistral AI: its terms for consumers in the European Economic Area, effective 8 October 2026, govern personal use. Under section 12.3 of its Commercial Terms (effective 25 September 2026), it may process personal data for a customer “as a Data Processor”, and its data processing agreement then applies. Section 4.2 allows training, among other cases, on feedback and on products set to training by default unless the customer opts out. Section 2.3 of the data processing agreement (effective 27 July 2026) names Mistral AI “as Controller” for such training and for automated moderation. According to its help centre (3 September 2026), administrators of its Team plan can switch training off, while Enterprise customers are opted out by default.
  • OpenAI: its Europe Terms of Use (updated 16 January 2026) apply to services for individuals. Its enterprise privacy page (updated 8 January 2026) says OpenAI can sign a data processing agreement for ChatGPT Business, ChatGPT Enterprise and the API, and states: “By default, we do not use your business data for training our models.” Data that a customer has explicitly opted in to share is an exception.

Is switching off training in the settings enough?

No: switching off training in a personal account limits one use of your inputs, but the provider remains the controller for its other purposes. A settings switch does not turn the account into processing on the practice’s behalf.

OpenAI’s Europe Terms of Use (updated 16 January 2026) offer a training opt-out in the account settings, but the general clause stays in force: “We can use your Content worldwide to provide, maintain, develop, and improve our Services”. Anthropic’s Privacy Policy (effective 10 September 2026) names Anthropic Ireland, Limited as controller for users in the European Economic Area. It allows training on inputs and outputs unless the user opts out; feedback and conversations flagged for safety review are exceptions.

For a practice, the test is whether the provider pursues its own purposes. The European Data Protection Board (EDPB) states that a party that in fact decides why and how personal data are processed “will be a controller even if a contract says that it is a processor” (Guidelines 07/2020, paragraph 29; see also Article 28(10) GDPR). The Hamburg data protection authority draws the practical line in its checklist of 13 November 2023: “If the terms and conditions allow the AI provider to process data for its own purposes, you should not transmit any personal data to the AI.”

Is a paid personal plan a business plan?

No: a paid plan for individuals is still a plan for individuals, and whether a data processing agreement applies depends on the type of plan. Anthropic’s Consumer Terms (effective 8 October 2025) state that its Commercial Terms, which incorporate its Data Processing Addendum, do not cover “Claude.ai or Claude Pro use for individuals or entities”. On that wording, an organisation that pays for Claude Pro gets no addendum through the Commercial Terms.

Business use does not change this. The addendum for “commercial or business use” in OpenAI’s Europe Terms of Use (updated 16 January 2026) deals with liability, indemnity and governing law. It contains no terms for processing personal data on the customer’s behalf.

Good to know: The Hamburg data protection authority advises organisations to provide their own accounts: employees “should not create an account on their own using private data”.

Which data belongs in which kind of AI service?

Under the GDPR, texts that contain no personal data, such as a notice about opening hours, can go into any kind of service, while personal data belongs only where the contract and a legal basis cover it. Personal data is any information relating to an identified or identifiable natural person (Article 4(1) GDPR), so leaving out the name is often not enough. Data concerning health is personal data “related to the physical or mental health of a natural person, including the provision of health care services, which reveal information about his or her health status” (Article 4(15) GDPR).

In a practice, this means:

  • Appointment emails, invoices and letters to named people: personal data, only with a data processing agreement or in a system the practice runs itself, and with a legal basis. In a health practice, they can also reveal data concerning health, such as the fact of a visit (Recital 35 GDPR).
  • Anything about a person’s physical or mental health: data concerning health. Article 9 GDPR prohibits processing it unless an exception applies.

Important: A data processing agreement sets out the provider’s duties as a processor; it does not replace the legal basis that the practice needs. If a provider in fact uses the data for its own purposes, such as training, the contract alone does not make it a processor.

What should a practice check before moving to a workspace?

Before the team’s first prompt in the workspace, the practice checks the contract and the accounts.

Fictional example: In a practice in Leipzig with two treatment rooms, the receptionist wants to tidy up an appointment email in a personal chatbot account and asks whether the patient’s name may stay in the text. The owner says no and books a business plan with a data processing agreement. As administrator, the owner sets up one account per team member. A written rule now lists what never goes into a prompt: names, dates of birth, contact details and anything about health.

  • The plan runs on business terms.
  • The contract includes a data processing agreement under Article 28(3) GDPR.
  • The provider offers sufficient assurance that it will implement appropriate technical and organisational measures (Article 28(1) GDPR).
  • The terms state whether training on inputs is off by default, which exceptions apply, for example feedback, and which purposes the agreement reserves for the provider.
  • An administrator of the practice manages the accounts, and the team knows who can see conversations.
  • Personal accounts stay out of practice work.
  • A written rule lists the details that never go into a prompt.

Frequently asked questions

Does ChatGPT Plus come with a data processing agreement?

The OpenAI pages linked in this guide do not offer one for the chats themselves: the pricing page, accessed on 6 November 2026, describes Plus as a plan designed to be used by individuals. OpenAI’s enterprise privacy page, updated on 8 January 2026, names ChatGPT Business, ChatGPT Enterprise and the API as the services for which OpenAI can sign a data processing agreement. A separate data processing addendum for ChatGPT Sites, published on 9 July 2026, covers Plus users only where a published ChatGPT Site collects personal data, and not their prompts or conversations.

Do I need a data processing agreement for every AI tool?

Only if the provider processes personal data on behalf of the practice. Article 28(3) GDPR requires a contract for that case, and whether the case arises depends on what goes into the tool and on the provider’s role. Drafting a practice flyer without any personal data does not trigger the requirement.

Can the practice see what the team types in a workspace?

It depends on the provider, the plan and its settings; for ChatGPT Business, OpenAI’s help article on admin controls for plugins and apps, updated on 5 November 2026, says administrators “do not automatically gain access to members’ ordinary private conversations”. OpenAI’s enterprise privacy page, updated on 8 January 2026, says of the same plan: “Workspace admins have control over workspaces and can view, access, export, and delete end user conversations in the workspace.” A further help article, updated on 3 November 2026, says that what administrators can access depends on the organisation’s configuration and applicable law, so check this before the team starts.

What happens to a personal account opened with a work email address?

It can come under the control of the organisation that owns the address. Under OpenAI’s Europe Terms of Use (updated 16 January 2026) and Anthropic’s Consumer Terms (effective 8 October 2025), such an account may be added to or linked with the organisation’s business account, and the organisation’s administrator may then access its content. OpenAI says it gives notice first, unless the organisation has already told the user that it may monitor and control the account; Anthropic says it gives notice before linking, but may skip it if the organisation is responsible for notifying the user or has already informed them.

Does a local AI system need a data processing agreement?

Only if an outside firm processes personal data for the practice: a model that the practice runs on its own hardware involves no outside AI provider, but the practice is then responsible for operation, updates and access control. The German data protection authorities (DSK) regard an organisation that runs an AI application on its own servers solely for its own purposes as the sole controller as a rule (guidance of 6 May 2024, paragraph 32). If an IT service provider runs the system, that service provider usually processes personal data on the practice’s behalf, which calls for a data processing agreement.

Sources

  1. Publications Office of the European Union: Regulation (EU) 2016/679 (General Data Protection Regulation), Official Journal L 119 of 4 May 2016 (eur-lex.europa.eu), accessed
  2. Federal Ministry of Justice, gesetze-im-internet.de: Heilpraktikergesetz (HeilprG), section 1 (www.gesetze-im-internet.de), accessed
  3. City of Munich (Landeshauptstadt München): Heilpraktiker licence in the field of psychotherapy (stadt.muenchen.de), accessed
  4. European Data Protection Board (EDPB): Guidelines 07/2020 on the concepts of controller and processor in the GDPR, version 2.1 (www.edpb.europa.eu), accessed
  5. Datenschutzkonferenz (DSK): Orientierungshilfe Künstliche Intelligenz und Datenschutz, Version 1.0, 6 May 2024 (www.datenschutzkonferenz-online.de), accessed
  6. Federal Office for Information Security (BSI): Artificial Intelligence (www.bsi.bund.de), accessed
  7. The Hamburg Commissioner for Data Protection and Freedom of Information: Checklist for the use of LLM-based chatbots, 13 November 2023 (datenschutz-hamburg.de), accessed
  8. Anthropic: Consumer Terms of Service, effective 8 October 2025 (www.anthropic.com), accessed
  9. Anthropic: Commercial Terms of Service, effective 17 June 2025 (www.anthropic.com), accessed
  10. Anthropic: Data Processing Addendum, effective 24 February 2025 (www.anthropic.com), accessed
  11. Anthropic: Privacy Policy, effective 10 September 2026 (www.anthropic.com), accessed
  12. Anthropic Privacy Center: Is my data used for model training? (commercial products) (privacy.claude.com), accessed
  13. Mistral AI: Terms of Service for consumers in the European Economic Area, effective 8 October 2026 (legal.mistral.ai), accessed
  14. Mistral AI: Commercial Terms of Service, effective 25 September 2026 (legal.mistral.ai), accessed
  15. Mistral AI: Data Processing Addendum, effective 27 July 2026 (legal.mistral.ai), accessed
  16. Mistral AI Help Center: Can I opt out of my input or output data being used for training?, 3 September 2026 (help.mistral.ai), accessed
  17. OpenAI: Europe Terms of Use, updated 16 January 2026 (openai.com), accessed
  18. OpenAI: Enterprise privacy at OpenAI, updated 8 January 2026 (openai.com), accessed
  19. OpenAI: Admin controls, security, and compliance for plugins and apps, updated 5 November 2026 (help.openai.com), accessed
  20. OpenAI: Data access for your managed ChatGPT account, updated 3 November 2026 (help.openai.com), accessed
  21. OpenAI: ChatGPT pricing (chatgpt.com), accessed
  22. OpenAI: ChatGPT Sites Data Processing Addendum, published 9 July 2026 (openai.com), accessed

ChatGPT is a trademark of OpenAI. Claude is a trademark of Anthropic.

About this guide

Publisher
sicherAI (see imprint)
Our interest
We offer an AI service for Heilpraktiker practices and use models from several makers, including some named in our guides. The guide contains no rankings. Statements about other providers rest on the sources listed.
How it was made
This article was created with AI.
Sources as of
Legal position as of
Limits
This guide provides general information. It is not legal advice for your individual case.
Report an error
Contact us using the details in our imprint.