Your data stays in the EU
The first question European companies ask about AI is almost always the same: where does our data end up? Our answer is clear. We deliver Claude via Amazon Bedrock in European AWS regions. Requests are processed exclusively within the EU, Frankfurt included. By default your content is not stored, is never used to train models, and is not shared with the model provider.
Requests run through Amazon Bedrock's EU inference profile and are processed exclusively in AWS regions in EU member states: Frankfurt, Ireland, Paris, Milan, Spain, Stockholm.
Neither AWS nor Anthropic trains models on your content. On Amazon Bedrock that is part of the service; at Anthropic it is contractually excluded in the commercial terms.
For the Claude models we deploy on EU projects, Amazon Bedrock does not store inputs or outputs by default. The model provider has no access to your prompts and responses.
Encryption in transit and at rest, optionally with your own KMS keys. Every request is logged in AWS CloudTrail together with the region that processed it, so your auditors can verify.
What the architecture looks like
Three steps, no exceptions:
Your CRM, Microsoft Teams or line-of-business app sends the request, encrypted, to an AWS endpoint in the EU, for example Frankfurt.
The model runs inside AWS infrastructure. The EU inference profile guarantees processing never leaves the EU geography. Traffic between regions stays on the AWS network and never crosses the public internet.
Inputs and outputs are not stored by default. CloudTrail records, for every request, which region processed it.
For projects with especially strict requirements we walk the architecture through with your data protection officer up front, including the data processing agreement and technical documentation.
What you can verify
Not promises: attestations and contracts you can check:
Amazon Bedrock is in scope of AWS's C5 attestation, the cloud criteria catalogue of Germany's Federal Office for Information Security. For many European IT departments the reference that counts.
Amazon Bedrock: ISO 27001, 27017, 27018, 27701 and SOC 1/2/3. Anthropic: ISO 27001, ISO/IEC 42001 for AI management systems, and SOC 2 Type II.
Processing under Art. 28 GDPR: the AWS data processing addendum with EU standard contractual clauses is automatically part of the AWS Service Terms. Anthropic likewise provides API customers a Data Processing Addendum with standard contractual clauses.
Since January 2026 AWS additionally operates the independent European Sovereign Cloud, first region in Brandenburg, Germany, with Amazon Bedrock among the initial services. For sovereignty requirements, an option we evaluate together with you.
Two paths, depending on your requirements
Not every project has the same constraints. We build both:
For companies with clear residency requirements. Processing exclusively in EU regions, no storage by default, no model-provider access to your content. Our recommendation for the DACH market.
For projects without a residency requirement. Contractually no training on your data, automatic deletion of inputs and outputs within 30 days, DPA with standard contractual clauses. Zero-data-retention arrangements are available for qualifying customers. Processing takes place in the US among other locations.
Honesty is part of the offer: not every newly released Claude model is immediately available with EU processing. We verify which models currently have EU inference profiles at project start and put it in writing.
What about the EU AI Act?
Since 2 February 2025, Article 4 of the EU AI Act applies: companies using AI must ensure a sufficient level of AI literacy among their staff, proportionate to their background and the context of use. The law does not require any specific certification or licence. Demonstrable measures such as training are the usual way to meet the duty.
We have prepared the AI Act for businesses, article by article in the original wording with practical notes. And if you want to address the literacy duty in a structured way, our Claude-certified architects train your team.
What this website itself does
Transparency is part of it. The AI assistant on this website currently runs on Anthropic's Claude API under its commercial terms: no training on your input, automatic deletion of inputs and outputs within 30 days. Please do not enter personal or confidential data there; a general description of your business is all the analysis needs.
Client solutions are built to the stricter standard above. Which architecture applies is your decision.
The questions companies ask us
No. With a Bedrock architecture using the EU inference profile, requests are processed exclusively in AWS regions in EU member states. Traffic between regions stays on the AWS network and is encrypted.
No. On Amazon Bedrock, Anthropic has no access to prompts and responses. And on the Claude API, Anthropic's commercial terms exclude training on customer content.
Not by default on Amazon Bedrock, for the Claude models we deploy on EU projects. On the Claude API, inputs and outputs are automatically deleted within 30 days; zero-data-retention arrangements are available for qualifying customers.
Yes. The AWS DPA under Art. 28 GDPR with EU standard contractual clauses is automatically part of the AWS Service Terms, and Anthropic provides API customers a Data Processing Addendum as well. We supply the technical documentation for your data protection officer.
Yes. Every request is logged in AWS CloudTrail together with its processing region. On top of that sit the attestations: BSI C5 and ISO 27001 at AWS for Amazon Bedrock, ISO 27001 and ISO/IEC 42001 at Anthropic.
No. Article 4 of the EU AI Act requires proportionate AI-literacy measures, such as training, but no specific certification or licence. We help you implement it pragmatically and verifiably.
Talk to us about your requirements
Tell us your data protection and data residency constraints. You get a clear architecture recommendation and the documentation for your data protection officer before any project starts.