--- title: "Claude EU Data Residency in 2026: What Anthropic Offers, and What European Teams Do Instead" description: "Anthropic does not currently offer EU data residency for Claude — every prompt is processed on US infrastructure and constitutes an EEA-to-US transfer." url: "https://sonomos.ai/blog/claude-eu-data-residency-2026/" source: "https://sonomos.ai/blog/claude-eu-data-residency-2026/" --- [Back to Blog](https://sonomos.ai/blog/) September 1, 202613 min read Claude Data Residency Europe GDPR AI Compliance # Claude EU Data Residency in 2026: What Anthropic Offers, and What European Teams Do Instead S [Sonomos Research](https://sonomos.ai/team/) · Sonomos The Sonomos research team writes about AI privacy, data protection, and how to use generative AI safely at work. **Short answer:** As of this writing, Anthropic does not offer EU data residency for Claude. All Claude inference runs on US-based infrastructure, on every tier — Claude.ai consumer accounts, Claude for Work (Team and Enterprise), and the Claude API. That means every prompt containing personal data of EU data subjects is an EEA-to-US transfer under Chapter V of the [GDPR](https://sonomos.ai/glossary/gdpr/), and no tier or region setting avoids it. It can be made lawful: Anthropic's [DPA](https://sonomos.ai/glossary/data-processing-agreement/) incorporates Standard Contractual Clauses, Anthropic participates in the EU-US Data Privacy Framework, and EDPB Recommendations 01/2020 describe the supplementary measures that close the Schrems II gap — the most effective being pseudonymisation on the device before the prompt ever leaves the EU. This guide covers what Anthropic actually offers in 2026, how it compares to OpenAI, Microsoft, and Google, and the mitigation stack European teams deploy instead of waiting for an EU region. One caveat first: residency offerings change quickly. Everything below reflects Anthropic's publicly documented posture as of this writing — verify it against Anthropic's current trust documentation and DPA before finalising a DPIA or procurement decision. ## What "EU data residency" actually means — and what it does not Data residency is often treated as a single yes/no checkbox. It is really three separate questions: - **Storage residency.** Where is data at rest — conversation history, uploaded files, logs — physically stored? The commitment most vendors offer first. - **Inference residency.** Where does the model actually run? This is the hard one. A vendor can store your history in Frankfurt while the prompt is processed by GPUs in Virginia — if inference happens in the US, the transfer happens, whatever the storage commitment says. - **Support and operations access.** Can US-based personnel access the data for support or abuse monitoring? Remote access from a third country is itself a transfer under the EDPB's interpretation. A second point that surprises many teams: **residency is not the same as GDPR compliance, and its absence is not automatic non-compliance.** The GDPR does not require personal data to stay in the EU; it requires a lawful mechanism for transfers out of the EEA (Chapter V) plus all the other obligations — lawful basis, transparency, Article 28 contracts, Article 32 security. Residency removes the transfer question; it answers none of the others. Our [Claude GDPR guide](https://sonomos.ai/blog/is-claude-gdpr-compliant-2026/) walks through the full analysis. ## Claude's residency posture in 2026, tier by tier The picture across Anthropic's tiers is unusually simple, because the answer is the same everywhere: | Claude tier | DPA available? | EU data residency | What that means in practice | | --- | --- | --- | --- | | Claude.ai Free / Pro | No | No | Consumer contract, US processing; not suitable for EU personal data in a work context at all | | Claude for Work (Team) | Yes | No | DPA with SCCs; all inference on US infrastructure | | Claude for Work (Enterprise) | Yes | No | Strongest contractual posture (audit logs, SSO, BAA available) — but still US inference | | Claude API | Yes | No | Zero-data-retention option for eligible customers; still US inference | Three implications follow: **"Claude Enterprise EU data residency" does not currently exist as a product option.** Teams searching for it are usually hoping the top-tier plan unlocks a region picker the way it does at some competitors. It does not, as of this writing. Enterprise buys a stronger contract — admin controls, audit logs, configurable retention — not a different geography. **The transfer analysis is mandatory, not optional.** With EU-only providers, a controller can sometimes architect the deployment so no Chapter V transfer occurs. With Claude, every prompt containing personal data is a transfer, so the DPIA and Article 30 record must document the mechanism (SCCs and/or DPF) and the supplementary measures every time. **One indirect route exists via the hyperscalers.** Claude models are also available through AWS Bedrock and Google Vertex AI, which offer their own regional infrastructure. Whether Claude is actually deployable in an EU region there — and what commitments attach to model invocation rather than storage — depends on each cloud's current model-availability matrix, which changes frequently. If regional pinning is a hard requirement, this is the avenue to investigate; verify it against AWS's and Google's current documentation rather than assuming. ## Is Claude European? A question people ask search engines directly: no, Claude is not a European product. Anthropic is Anthropic PBC, a US-headquartered public benefit corporation, and its disclosed sub-processors are led by US cloud infrastructure — AWS primarily — plus support tooling. Anthropic serves European customers, offers an EU-transfer-ready DPA, and participates in the EU-US Data Privacy Framework, but from a GDPR standpoint it is a US data importer, with everything Chapter V implies. That is not a disqualifier — OpenAI and Google sit in the same position — but it is the correct starting fact for any European risk assessment. ## How Claude compares: residency options across the major providers Here is where the major providers stood on EU residency as of this writing. These offerings change often — treat the table as a snapshot and verify each vendor's current documentation before relying on it. | Provider / product | EU data residency | Notes | | --- | --- | --- | | Claude (all Anthropic tiers) | No | All inference on US infrastructure; SCCs + DPF for transfers | | ChatGPT Enterprise | Partial | EU residency for eligible customers covers stored content; inference compute may still transfer | | OpenAI API | Partial | EU residency available on eligible plans; verify current scope | | Azure OpenAI Service | Yes, by region selection | European customers can select EU Azure regions | | Microsoft 365 Copilot | Yes (EU Data Boundary) | Prompts and responses processed and stored in EU/EEA for EU tenants | | Google Workspace Gemini | Partial | EU residency for stored Workspace data; AI inference may involve global infrastructure | | Vertex AI Gemini | Yes, for eligible workloads | EU-only inference available on eligible Vertex AI workloads | The pattern is clear: the hyperscaler-hosted routes (Azure, Vertex, the EU Data Boundary) are where real regional pinning lives today. Anthropic is the one provider in this list with no first-party residency option at all as of this writing — which is why the mitigation stack below matters more for Claude deployments than for most. ## Anthropic and the EU-US Data Privacy Framework Anthropic participates in the EU-US Data Privacy Framework (DPF), the adequacy arrangement in force since July 2023. This gives EU controllers a second lawful transfer mechanism alongside SCCs, and the two work differently: - **DPF (Article 45).** The European Commission has decided that transfers to certified US organisations are adequately protected. If Anthropic's certification covers the relevant data, the transfer is lawful without SCCs and, formally, without a Transfer Impact Assessment. - **SCCs (Article 46).** The 2021 Standard Contractual Clauses — incorporated in Anthropic's DPA, Module 2 (controller-to-processor) or Module 3 where relevant — plus a documented TIA addressing US surveillance law. Most European procurement teams rely on **both**: DPF as the primary mechanism, SCCs as the fallback. The reason is history. Schrems I ended Safe Harbor in 2015; Schrems II ended Privacy Shield in 2020; the DPF is the third attempt at US adequacy, a Schrems III challenge remains a live possibility, and several EU supervisory authorities still voice concerns about US government access under FISA 702. If the DPF falls, contracts that relied on it exclusively lose their transfer basis overnight; contracts with SCCs already incorporated keep working. Neither mechanism changes what physically happens: the data goes to US infrastructure, within reach of US legal process. That is why EDPB Recommendations 01/2020 call for **supplementary measures** on top of the legal mechanism — and why the most robust measure is technical: transform the personal data before it leaves the EU, so that what crosses the Atlantic is no longer identifiable. ## What goes wrong: the residency failure patterns The recurring mistakes in European Claude deployments, specifically on the residency and transfer question: 1. **Assuming Enterprise means EU.** A procurement team checks "data residency: yes" because the Enterprise tier has enterprise-grade everything else. The DPIA then documents a transfer that supposedly does not happen, and the whole assessment loses credibility under audit. 2. **Relying on the DPF alone.** The DPA is signed with DPF as the only referenced mechanism — no SCC fallback, no TIA. Lawful today, fragile against a Schrems III ruling, and out of step with what supervisory authorities expect to see documented. 3. **Ignoring the transfer because "it is just prompts."** A prompt containing a customer name, an employee grievance, or a patient detail is personal data in transit to a third country. Volume and informality do not change the Chapter V analysis. 4. **No supplementary measures at all.** SCCs signed, TIA filed, and then raw personal data flows to US infrastructure in every prompt. EDPB Recommendations 01/2020 exist precisely because the clauses alone cannot bind a foreign government; the paper-only transfer file is the pattern regulators have been least sympathetic to since Schrems II. 5. **Consumer Claude accounts under the radar.** Employees use Claude.ai Pro with work data: a transfer with no DPA, no SCCs, no enterprise DPF coverage — an Article 28 breach and an unlawful transfer in one motion. ## Claude and the EU AI Act Two clarifications for teams conflating the residency question with the [EU AI Act](https://sonomos.ai/glossary/eu-ai-act/): **Residency is not an AI Act concept.** The AI Act regulates what AI systems do — risk categories, transparency, human oversight — not where the servers sit. An EU-resident deployment can violate the Act; a US-processed one can comply. The Act applies extraterritorially wherever the system's output is used in the EU. **Anthropic carries the GPAI provider obligations; you carry the deployer ones.** As a general-purpose AI provider, Anthropic is responsible for the GPAI obligations applicable since August 2025 — technical documentation, training-data transparency, copyright policy, incident reporting — and deployers should collect that documentation as vendor due-diligence evidence. Deployer obligations attach to you: Article 50 transparency, and the full Annex III high-risk stack from August 2026 if Claude is a substantial component in employment, credit, education, or other high-risk decisions. Our [EU AI Act compliance checklist](https://sonomos.ai/blog/eu-ai-act-compliance-checklist-2026/) covers those obligations phase by phase. ## The practical alternative: minimise before you transfer Since the transfer to US infrastructure cannot currently be avoided with Claude, the highest-leverage control is to change **what** transfers. EDPB Recommendations 01/2020 identify pseudonymisation before transfer — with the means of re-identification held only in the EU — as an effective supplementary measure: the US importer, and any government request served on it, sees data that cannot be linked back to individuals. This is the layer where Sonomos Locke sits. Locke 1.0 is a local-first privacy layer that runs in the browser: it detects names, identifiers, contact details, health terms, and account numbers on-device and replaces them with one-way placeholders before the prompt leaves the machine. Because detection happens locally, the original sensitive content never reaches Anthropic's infrastructure — or Sonomos. Reversible tokenization is planned shortly after 1.0. The effect on the transfer analysis is direct. If [pseudonymisation](https://sonomos.ai/glossary/pseudonymization/) happens before the prompt leaves EU-controlled devices, what crosses the Atlantic is pseudonymous — a materially lower-risk category under the EDPB framework — and where the redaction renders the data effectively anonymous, Recital 26 takes it outside the GDPR's scope entirely. It also satisfies the Article 32(1)(a) technical-measure expectation and strengthens the [data-minimisation](https://sonomos.ai/glossary/data-minimization/) position under Article 5(1)(c). To be precise: a masking layer is a supplementary measure and an Article 32 control — it does not create EU data residency, and it does not by itself make a deployment compliant; lawful basis, the DPA, the Article 30 record, and the DPIA remain the controller's work. What it does is convert an unavoidable transfer of identifiable personal data into a documented, defensible transfer of minimised data — exactly the posture EDPB guidance describes. ## What to check in Anthropic's DPA and trust documentation Before signing, and at each annual review, verify the following against Anthropic's current DPA, trust documentation, and sub-processor disclosures — not against blog posts, including this one: - **Residency status.** Has an EU data residency or regional processing option been introduced since this was written? Vendors add regions without fanfare; check before documenting its absence in a DPIA. - **Transfer mechanism.** Confirm the DPA incorporates the 2021 SCCs (the correct module for your relationship) and that Anthropic's DPF certification is current and covers the relevant data categories. - **Retention terms per tier.** What is the default retention for prompts and outputs, is it configurable, and does your workload qualify for the API zero-data-retention option? - **Sub-processor list.** Review the current list (cloud infrastructure — AWS primarily — plus support tooling), confirm processing locations, and confirm you receive advance notice of additions as Article 28(2) requires. - **Training exclusion.** Confirm prompts and outputs are excluded from model training on your tier, contractually and not just as a settings default. - **Support-access geography.** Ask where support and abuse-monitoring personnel sit and under what controls they can access customer content. - **GPAI documentation.** Collect Anthropic's EU AI Act GPAI documentation for your vendor file. Document each check with a date. A transfer file showing annual re-verification is worth far more under supervisory scrutiny than one assembled once at signature. ## The bottom line Anthropic offers no EU data residency for Claude as of this writing — every prompt from a European team is processed on US infrastructure, and every prompt containing personal data is a Chapter V transfer. That is a real gap against Azure OpenAI, M365 Copilot, and Vertex AI, and it should be named explicitly in the DPIA rather than papered over. But it is a manageable gap, not a disqualifying one. The lawful mechanisms exist — Anthropic's DPA with SCCs, plus its DPF participation — and the supplementary measure that does the real work is architectural: pseudonymise personal data on the device, inside the EU, before the prompt leaves. Claude for Work or the API under a signed DPA, a local-first masking layer like Locke, a documented TIA with SCC fallback, and an annually re-verified transfer file is a posture that holds up — with or without an EU region on Anthropic's roadmap. If Anthropic ships one, teams that built the minimisation layer will simply have one less residual risk, not a program to rebuild. ## Related reading - [Is Claude GDPR Compliant? A 2026 Guide for European Teams](https://sonomos.ai/blog/is-claude-gdpr-compliant-2026/) - [GDPR and AI: A 2026 Compliance Guide for European Teams](https://sonomos.ai/blog/gdpr-ai-compliance-eu-chatgpt-claude-gemini-2026/) - [EU AI Act Compliance Checklist for Enterprise Deployers](https://sonomos.ai/blog/eu-ai-act-compliance-checklist-2026/) - [Is Gemini GDPR Compliant?](https://sonomos.ai/blog/is-gemini-gdpr-compliant-2026/) - [Is Microsoft Copilot GDPR Compliant?](https://sonomos.ai/blog/is-microsoft-copilot-gdpr-compliant-2026/) - [ChatGPT vs Claude for Confidential Work](https://sonomos.ai/blog/chatgpt-vs-claude-confidential-work-2026/) ### Protect your data while using AI Locke detects and masks sensitive information on your device, before it reaches an AI model. [Download Locke](https://sonomos.ai/downloads/) This article is general information about privacy and AI, not legal advice, and reading it does not create a solicitor–client or attorney–client relationship. Regulatory requirements turn on your specific facts and jurisdiction and change over time. Consult qualified counsel before relying on anything here, and check the publication and last-reviewed dates above. We may revise posts, but we do not undertake to keep any post current as the law changes. ## Reference pages for this post - [Locke for ClaudeTool guide](https://sonomos.ai/for/claude/) - [GDPRGlossary](https://sonomos.ai/glossary/gdpr/) - [Data ResidencyGlossary](https://sonomos.ai/glossary/data-residency/) ## Related Articles September 1, 2026•Sonomos Research ### The Anthropic DPA Explained: Claude Data Processing Agreement Guide (2026) Anthropic offers a DPA on Claude Team, Enterprise, and the API — not on Claude.ai Free or Pro. Here is what the Anthropic DPA covers, how it handles EU-US transfers via SCCs and the DPF, what to verify before signing, and where a DPA stops. Claude DPA GDPR +2 September 1, 2026•Sonomos Research ### Can My Employer See My Claude Chats? What's Visible in 2026 (and What Isn't) Whether your employer can see your Claude chats depends on the account (personal Claude.ai vs. work-managed Team/Enterprise), the device, and the network. Here is what is visible in each scenario — including Claude Code — and what deleting a chat actually does. Claude Workplace Privacy AI Privacy +1 September 1, 2026•Sonomos Research ### Can My Employer See My Gemini Chats? Google Workspace Visibility Explained (2026) Whether your employer can see your Gemini chats depends on the Google account, the device, and the network. A Workspace-managed account sits inside your employer's estate; company Wi-Fi alone cannot read chat content. Here is exactly what is visible in each setup. Gemini Google Workspace Workplace Privacy +1