AI providers’ terms
What the assistant fetches from Fortnox becomes part of the chat with your AI provider, and your agreement with them applies there. Here is what the plans promise, and where to change the setting.
In short
If you use ChatGPT, Claude or Grok in your company’s name, on a business plan, the provider does not train on your content and you can get a data processing agreement. With a personal account a setting decides whether chats may be used for training, and you should check it before you ask about your books. The terms for Copilot and for the Gemini API, which Gemini CLI uses, are further down.
We send something to the AI provider only in answer to a question in the chat. Sometimes a little more comes along than what you asked about, such as what has happened since last time. If some areas should never reach the assistant at all, turn them off with protection mode on your connection page.
What the providers themselves say.
| ChatGPT Business, Enterprise and Edu | Content is not used to train OpenAI’s models, nor is what ChatGPT fetches through apps and connectors. OpenAI signs a data processing addendum for Business and Enterprise. OpenAI on business plans (updated ) |
|---|---|
| ChatGPT Free, Plus and Pro | OpenAI may use chats to train its models. Turn it off under Settings → Data controls with Improve the model for everyone. If you rate a response (thumbs up or down), that whole conversation may still be used. OpenAI on training (updated ) |
| Claude Team and Enterprise | Anthropic does not use the content for training unless you share it yourselves, for example by rating a response. The data processing addendum is part of Anthropic’s commercial terms automatically. Anthropic on training (updated ) · Anthropic on the DPA (updated ) |
| Claude Free, Pro and Max | You choose in Claude’s Privacy Settings whether chats may be used for training. If it is on, they are kept for up to five years, otherwise for 30 days. Anthropic on consumer terms (published ) |
| Microsoft Copilot with a work account | When an organisation uses Microsoft Copilot, Microsoft’s Data Protection Addendum and Product Terms apply, with Microsoft as processor, and chats are not used to train the foundation models. Web searches Copilot makes are not covered by the addendum. Microsoft on data protection in Copilot (updated ) |
| Grok Business and Enterprise | xAI, which calls itself SpaceXAI on its own pages, does not train its models on business content, neither prompts nor answers. If xAI processes personal data for you, its data processing addendum is part of the enterprise terms automatically. xAI on business plans (updated ) · xAI’s data processing addendum (effective ) |
| Grok with a personal account, on grok.com and in the app | xAI may use chats to train its models. Turn it off on grok.com under Settings → Data with Improve the Model, and in the app under Settings → Data Controls. It applies to new chats. Chats in Private Chat are never used for training. If you give feedback on a response, it may still be used. xAI on training for individuals (updated ) |
| Gemini API, for example Gemini CLI with a key from Google AI Studio | For users in the EEA, Switzerland and the UK, Google’s terms for paid services also apply to the free tier: prompts and responses are not used to improve Google’s products. Google keeps them for a limited time, only to detect and prevent abuse. Google’s Gemini API terms (updated ) |
| The Gemini app with a personal account, on gemini.google.com | Custom connectors require Keep Activity to be on, and then Google uses the activity to develop and improve its services, including training AI models, and some chats are reviewed by people. With the setting off, the models are not trained on the chats, but custom connectors cannot be used. Custom connectors are for now only available in the US. Google on privacy in the Gemini app (updated ) · Google on custom connectors (checked ) |
Checked against the providers’ own pages on , Google’s on . The providers change their terms, so read their pages before you decide.
Who is responsible for what
You are the controller of the data in your bookkeeping and choose which AI app and which plan you use.
We are a processor for what passes through the service when you ask. We keep no copy of the bookkeeping, and the AI provider is not our sub-processor. Read more in the data processing terms and about the sub-processors.
The AI provider processes the chat, with what the assistant fetched, under your agreement with them. If your bookkeeping holds sensitive data, patient names for example, your data protection officer should assess the plan and whether to use protection mode.
Questions about the terms?
We answer written questions about how data is handled with us. The AI provider answers about its own terms.