AI and privacy
For each Claude plan: what happens to your input, how long it is kept, whether an agreement is in place and where your data is stored. October 2026.
That depends on the plan and what you enter. With Claude Team, Claude Enterprise and the API, your input is not used for training by default and a data processing agreement applies. With Free, Pro and Max, it does not. On every plan, data is stored in the United States and everything an employee puts in the window goes along.
This article explores one question; the full overview is on AI chatbots and the GDPR.
“Safe” here is not a property of Claude AI itself. It concerns four questions the GDPR asks of every supplier: what does it do with input, how long does it retain it, under which contract, and where? This article checks those questions for each plan, based on Anthropic's privacy centre, terms and pricing page. This is the position on 13 October 2026. These pages change several times a year: open the links before recording a decision.
This is an explanation, not legal advice. The same comparison for ChatGPT appears in ChatGPT Free, Plus, Business or Enterprise: which plan may you use for business?. The broader question of what goes wrong when someone uploads a case file is on ChatGPT, Claude, Mistral Vibe and the GDPR.
Claude is made by Anthropic. On its company page, it describes itself as an AI safety and research company and has the form of a Public Benefit Corporation. The same page says services in the EU are supplied by Anthropic Ireland Limited.
Under Anthropic's privacy policy, effective since 10 September 2026, Anthropic Ireland is the controller for users in the EEA, the United Kingdom and Switzerland. An Irish contracting party does not change where the data is stored.
Anthropic has two sets of terms, and each plan falls under one of them:
The Claude pricing page lists six options, as at October 2026:
We leave out prices: the pricing page shows dollar amounts, and Anthropic says they differ by region. The dividing line between Max and Team matters more. Moving from Pro to Max buys more usage. Moving from Max to Team gives you a different contract.
According to Anthropic's privacy centre, conversations on consumer plans are used to improve models if the user opts in. The setting is called “Model Improvement” and appears in the privacy settings. The person holding the account makes the choice. As an employer, you cannot see its state.
Anthropic names three exceptions:
For connections, the same page says raw content from connectors, including MCP servers, does not enter training data unless copied directly into the conversation.
The page about consumer retention periods lists these periods, as at October 2026:
A conversation nobody deletes remains stored.
The consumer terms for the EEA and Switzerland say users do not use the services for commercial or business purposes. Doing client work on a Pro account therefore falls outside what those terms envisage. There is no data processing agreement, and the organisation has no administrative control over the account. See Shadow AI: what it is, how serious the risk is and how to control it.
Claude Team is the lightest plan where the organisation is the customer. You purchase it yourself, for at least two people.
For commercial products, Anthropic says conversations are not used to train models by default. Exceptions include participation in a development partner programme and material someone expressly reports, for example with a thumbs up or down. Agree that nobody uses those buttons during confidential work.
You do not need to request the data processing agreement. Anthropic says it automatically forms part of the commercial terms, including standard contractual clauses. The agreement itself includes:
Read the agreement yourself and record the version. What to check is covered in Data processing agreement with OpenAI, Anthropic, Microsoft or Mistral: what should you check?.
The page about organisational retention periods says conversations remain so you can continue them. Delete a conversation and it leaves underlying systems within thirty days. Team does not allow custom retention periods: according to the pricing page, these belong to Enterprise.
For administration, the pricing page lists SSO sign-in, control over permitted connectors and usage statistics for Team.
The pricing page lists Enterprise features missing from Team: audit logs, SCIM for account management, role-based access, an oversight connection and custom retention periods. The last feature works as follows: an organisation owner sets a retention period of at least thirty days. If nobody sets anything, data is retained indefinitely. An Enterprise contract without a configured period therefore keeps data longer than you might expect.
Anthropic says API input and output are deleted within thirty days of receipt, with exceptions for features where you choose storage, enforcement of usage policies and legal obligations. A no-retention arrangement exists only if separately agreed with Anthropic. The API falls under the same commercial terms and data processing agreement.
If you obtain Claude through another party's platform, Anthropic says that platform's terms apply. You then put this article's questions to that party.
As at October 2026, based on the Anthropic pages linked above. Check them before deciding.
| Plan | Training on input | Retention period | Data processing agreement | Storage location | Administration |
|---|---|---|---|---|---|
| Free | User chooses; feedback and flagged conversations may always be used | Until the user deletes, then removed within 30 days; up to 5 years if training is allowed | No | United States | None |
| Pro | As Free | As Free | No | United States | None |
| Max | As Free | As Free | No | United States | None |
| Team | Not by default; feedback may be used | Until deletion, then removed within 30 days | Yes, part of the commercial terms | United States | SSO, connector management, statistics; 2 to 150 seats |
| Enterprise | Not by default; feedback may be used | Custom period from 30 days; indefinite without a setting | Yes, part of the commercial terms | United States; US-only processing can be configured | SSO, SCIM, audit logs, roles |
| API | Not by default | Removed within 30 days, with exceptions | Yes, part of the commercial terms | Storage in the United States | Workspaces and keys |
On its page about server locations, Anthropic says data is stored in the United States. Customer traffic may by default be processed in several countries across the US, Europe, Asia and Australia.
Your settings are limited. Enterprise customers with usage-based billing can restrict processing to the United States. For the API, the data residency documentation describes two processing choices, global or US-only, and one storage region: the United States. No European region is listed.
Chapter V of the GDPR allows transfers to a country outside the EEA only subject to conditions: a European Commission adequacy decision (Article 45) or appropriate safeguards such as standard contractual clauses (Article 46). Anthropic's privacy policy names both routes. The data processing agreement contains standard contractual clauses.
This provides a mechanism. The justification is yours: what data goes there, why is it necessary, and what risks do you identify? Record it in your record of processing activities and, where needed, a DPIA. See DPIA for AI tools: when is one mandatory and how do you approach it?.
A business plan arranges the relationship with Anthropic. Two things remain, alongside your own lawful basis and who within your organisation may see each case file.
On every plan, data is stored in the United States. If you do not want that or cannot justify it, a more expensive Claude plan does not help. You then arrive at a provider storing data in the EU, or a process where personal data does not go along. What a European provider solves and what it does not is covered in European AI: alternatives to ChatGPT at a glance.
The plan does not check whether a file contains a BSN, diagnosis or salary. Uploading a forty-page case file to find one date sends thirty-nine pages too many. Article 5(1)(c) GDPR requires you to process only what is necessary. See Uploading documents into ChatGPT or Claude: what happens to them?.
PrivaSearch does not replace a Claude subscription. It addresses the remaining issue described above: what is entered. Instead of uploading a case file, you let the assistant search passages from which personal data has already been removed.
This happens in a fixed order. Documents are analysed inside PrivaSearch, without an external AI provider. Detected data such as names, BSNs (Dutch citizen service numbers) and addresses are replaced with pseudonyms that remain consistent within a project. A colleague checks the result alongside the original and approves it. Until then, a document is not searchable. See How PrivaSearch works.
The assistant connection uses MCP, the open standard through which an assistant uses external tools. You can connect Claude (in the browser, desktop app and Claude Code), as well as ChatGPT Business and Mistral Vibe (formerly Le Chat). The connection uses OAuth 2.1 and access tokens expire after eight hours. You connect at your organisation's own address, approve the connection after signing in and can revoke it at any time on your settings page. For each search, the assistant receives at most eight released passages with a short source reference, and no files. MCP is explained in What is an MCP server?.
Two things remain your responsibility. What PrivaSearch returns becomes part of the conversation and reaches the assistant's provider, under its retention policy. What you type yourself goes straight to the provider: ask about “the tenant”, rather than a name. We therefore recommend a business account with a data processing agreement for the connection. Pseudonymised data remains personal data for your own organisation, so the record and DPIA remain your responsibility. Also see The risks of AI for your client files.
To see Claude answer a question about a case file without the file going along, request a demo.
That depends on the plan. On Claude Team, Claude Enterprise and the API, Anthropic does not train on your input by default and a data processing agreement applies. On Free, Pro and Max, the user chooses about training and there is no agreement. With every plan, data is stored in the United States. No plan regulates what staff enter: you agree that yourself.
Claude Pro is no basis for client or employee data. It is a personal plan under consumer terms, which say that users in the EEA do not use the service for commercial or business purposes. There is no data processing agreement, and your organisation has no administrative control over the account. Work with personal data starts at Claude Team.
Both fall under commercial terms, include the data processing agreement and do not train on input by default. Claude Team is for 2 to 150 people and has SSO and connector management. Claude Enterprise adds audit logs, role-based access, account management through SCIM and custom retention periods. This is the position in October 2026 according to Anthropic's pricing page.
On Free, Pro and Max, only if you allow it with “Model Improvement”, give feedback or a conversation is flagged for safety review. On Team, Enterprise and the API, it does not happen by default. There too, material you submit yourself, such as a thumbs up or down, may be used for training. This is the position in October 2026.
No. Anthropic's privacy centre says data is stored in the United States. Traffic may by default be processed across several regions, including Europe, but that is not European storage. Enterprise customers and API users can restrict processing to the United States. October 2026 documentation lists no choice of a European region.
No. Claude Team arranges the agreements with Anthropic: no training by default and a data processing agreement. The rest remains your responsibility: a lawful basis, justification for transfers to the United States, an entry in your record, a DPIA where necessary, and agreements about what staff enter. No product arranges this for you. If in doubt, involve a lawyer or your data protection officer.
What you need to arrange for each type of data before using an AI chatbot: from names and email addresses to health, criminal records and BSNs.
Where an uploaded file ends up, how long it stays, who can access it and what deletion means. With four approaches for files containing personal data.
How a healthcare practice or healthcare administration uses PrivaSearch and its own AI assistant to search intake reports, care contracts and complaint files.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →