Claude in business: what you need to know before using it

Alongside ChatGPT, Claude from the US provider Anthropic is one of the most widely used AI assistants in German companies, precisely because of its strengths with long documents and texts. Many freelancers and small teams simply keep working with the personal account they already had. That is exactly where the problem begins: whether your inputs are used for training, whether you get a data processing agreement (DPA) and how long your data is stored depends entirely on which type of access you use and how it is configured.

How does Claude handle your data?

Since a change to its terms of use in autumn 2025, Anthropic uses chats from the personal plans Free, Pro and Max to train its models unless you opt out. The relevant toggle was already switched on in the consent pop-up, and many people simply clicked through. The consequences are real: with training switched on, Anthropic stores your chats for up to five years. If you switch it off, it is 30 days.

You can check this in the claude.ai settings, in the Privacy section, under "Help improve Claude". Important: the toggle only applies to future chats. Anything that has already gone into training cannot be taken back.

Personal account or business access? The most important decision

Even before any settings, your choice of access determines whether Claude is fit for business use. Free, Pro and Max are covered by the Consumer Terms: no data processing agreement, possible training, no central administration. Team, Enterprise and the API are covered by the Commercial Terms: Anthropic does not train on your content by default, and the data processing agreement is automatically part of the contract.

A note, because contradictory information is circulating on this point: at Anthropic, the Team plan is also covered by the Commercial Terms, including the data processing agreement. Anyone who tells you that only the Enterprise plan is fit for business use is wrong.

Settings you should check

• Training toggle: if you are (still) using a personal account, switch off "Help improve Claude" in the privacy settings.
• Memory: Claude now remembers facts and preferences across chats by default. Look through the saved entries and delete anything that does not belong there. The feature can be switched off completely.
• Incognito chats: for sensitive one-off questions, there are chats that end up neither in your history nor in memory and are not used for training.
• Integrations: connections to Google Drive, your calendar or email give Claude access to entire data sets. Only connect what you really need.
• Shared links: anyone with the link can read shared chats and artifacts. Before sharing, check whether they contain customer data.

What can go into a prompt?

Personal data such as names, addresses or customer enquiries that show the sender only belongs in Claude if a data processing agreement is in place, so only in the business versions, and even there only as far as the purpose requires. It has no place in personal accounts. The same applies to trade secrets: pricing calculations, contracts and source code only go into the approved environment, never into a free personal account. The simplest approach is one basic rule: anything you would not also send to an external service provider without a contract stays out.

Special case: professional secrecy

Lawyers, tax advisers, doctors and other professionals bound by professional secrecy face an additional hurdle: Section 203 of the German Criminal Code (StGB). The standard plans of the major AI providers do not include a confidentiality undertaking tailored to it. Client and patient data therefore only belongs in the tool in anonymised form, regardless of the plan you choose. How law firms can assess their use of AI step by step under Section 43e of the Federal Lawyers' Act (BRAO) is explained in the article AI in law firms: what is allowed.

Where does your data end up?

According to Anthropic, data is stored in the US, even if some requests are routed through European infrastructure. For European customers, the contracting party is Anthropic's Irish company, and the transfer to the US is based on the EU standard contractual clauses included in the data processing agreement. That is a workable route, but you should document it properly in your record of processing activities.

If you have stricter requirements, you can use Claude models through cloud providers such as AWS or Google Cloud in European data centres, for example in Frankfurt. Processing then stays in the EU. Anthropic itself is certified to SOC 2, ISO 27001 and ISO 42001. However, these certificates cover the provider's infrastructure, not how you use the tool.

What obligations apply to your business?

• Have a DPA in place: included automatically with Team, Enterprise and the API. Archive the version of the agreement for your records.
• Update your record of processing activities: the AI assistant is a processing activity in its own right.
• Update your privacy policy if customer data runs through Claude, for example with automated replies in customer support.
• Train your team: the obligation to ensure AI literacy under Article 4 of the EU AI Act has applied since February 2025.
• Label AI: since 2 August 2026, customers must be able to tell when they are chatting with an AI (Article 50 EU AI Act).
• Set rules: a short AI policy sets out who may use Claude, and for what. Our guide to AI policies shows how to do it.

How to set up Claude properly with Klaus

With Klaus, all the points in this article come together in one guided process: in the onboarding chat, Klaus adds Claude to your AI inventory as a tool profile, with its purpose, type of use and data categories. Along the way, he automatically asks exactly the questions that matter here: does Claude run via the app, the API or a cloud region? What data may go in? Are you subject to a duty of confidentiality under Section 203 StGB? From your answers, Klaus creates the classification of the tool, your AI policy as a PDF and training that builds its examples on your real tools, Claude included.

And because the question of server location plays a central role above: we take the route via European cloud infrastructure described there ourselves. Klaus runs on Claude models via AWS Bedrock in an EU region. So helping you with compliance does not mean sending the information from your check-up to the US.

Finally: the latest developments

Two developments from July 2026 complete the picture. First, the $1.5 billion copyright settlement between Anthropic and US authors received final court approval. It concerns past training data, not your use, but it shows how seriously the legal questions around AI training are being taken. Second, Anthropic itself still does not offer data storage in the EU. If you need that, take the route via the European cloud regions of AWS or Google.

Conclusion

Claude can be used properly in your business if the basics are right: business access with a data processing agreement, training switched off, a quick look at memory and integrations, and clear rules for the team. Once you have taken care of these points, you get one of the strongest AI assistants on the market without leaving yourself exposed on compliance.