Claude Enterprise security: SOC 2, ISO and GDPR
Most Claude rollouts we run in Europe meet one checkpoint before the first licence is bought, when the security lead asks for proof that Anthropic deserves the files staff will paste in. Gut feeling usually answers that request, while the audit reports that would settle it sit unread in a trust portal.
The guide below reads Claude Enterprise security straight from Anthropic's Trust Center and pricing page as they stood on 4 November 2026.12 Which plan suits which team sits in our guide to Claude for business, so this one ends with your own tasks.
Is Claude secure enough for a European business?
Measured against what a security review in a European company normally asks for, the short answer is yes. Anthropic's Trust Center marks both Claude Team and Claude Enterprise as covered by three independent audits, which the next section explains one by one.1 The Data Processing Addendum, or DPA, meaning the contract that makes Anthropic your processor under the GDPR, is part of the Commercial Terms that EU customers sign with Anthropic Ireland, Limited.3 Commercial inputs and outputs are not used to train Anthropic's models by default, which is the line a data protection officer usually asks to see in writing.4
The real gap sits between the two business plans, because Claude Team leaves out audit logs, automatic account provisioning and custom data retention, which only Claude Enterprise carries.2 Certificates also say nothing about what staff paste into a chat on a Tuesday morning, so the risk left after purchase sits in your settings.
Claude Enterprise security in numbers: SOC 2, ISO 27001 and ISO 42001
Three audits carry most of a vendor review, and the Trust Center lists all three for Claude Team, Claude Enterprise and the API, with reports for customers who request access.1
Claude and SOC 2 Type II
SOC 2 Type II is an auditor's report on whether a company's security controls actually worked across a whole audit period, not only whether they were designed well on a single day.5 Anthropic offers the 2025 report and two 2026 bridge letters, meaning letters from Anthropic covering the months since the audited period ended.1
Claude and ISO 27001:2022
ISO 27001 certifies the system a company uses to run information security, meaning the policies, risk reviews and controls that an accredited certification body audits. Anthropic holds ISO 27001:2022, the current version, with a 2025 certificate in its Trust Center.16 Put a copy in your vendor file now, because the security questionnaires your own customers send will ask for it by name.
Claude and ISO/IEC 42001:2023
ISO/IEC 42001 is what Anthropic calls the first international standard for AI governance, meaning a framework for finding, assessing and reducing the risks of AI systems.7 Anthropic announced its certification on 13 January 2025, issued by the certification body Schellman Compliance, and described itself as one of the first frontier AI labs to reach it.7 Quote this certificate when a tender asks for documented AI governance, since a SOC 2 report says little about how a model behaves.
Where is Claude data stored, and what does it mean for GDPR?
Anthropic states that data from its commercial products is stored in the US, while customer traffic may by default be routed to select countries on four continents, Europe among them.8 No EU-only processing exists on Anthropic's own platform, since the API runs requests only globally or in the US, and Claude Team has no location setting.910
Transfers to the US rest on the Standard Contractual Clauses, meaning the European Commission's model contract for sending personal data abroad, which Anthropic's DPA includes under Irish law.311 Our walkthrough of the Anthropic DPA shows whether Module Two or Module Three fits your setup.
A business that needs processing inside the EU picks a cloud route instead: EU cross-region inference on Amazon Bedrock, meaning requests that stay inside EU regions, or the EU multi-region endpoint on Google Cloud.1213 Both are API products for developers rather than a chat window for thirty staff, and our Claude EU data residency guide compares them model by model.
Residency answers where a request runs, while the GDPR still asks whether the transfer is lawful, and our answer to whether Claude is GDPR compliant lists the eight checks that settle it.
Claude Team vs Claude Enterprise: which security controls you get
Certificates prove the vendor is audited, while controls are what your own admins switch on, and here two plans with the same $20 seat price differ by seven lines.2
| Control | What it does | Claude Team | Claude Enterprise |
|---|---|---|---|
| Single sign-on (SSO) | Staff log in with their company account | Yes | Yes |
| Domain capture | Pulls every account on your domain into the workspace | No | Yes |
| SCIM provisioning | Adds and removes accounts from your staff directory | No | Yes |
| Role-based access | Sets permissions per role | No | Yes |
| Audit logs | Records who did what, exportable for 180 days | No | Yes |
| Custom data retention | Deletes chats after a period you set, from 30 days | No | Yes |
| IP allowlisting | Admits only network addresses you approve | No | Yes |
| Invoicing with payment terms | Replaces the card with an invoice | No | Through Anthropic sales |
Claude Team security settings for day one
Tick Require SSO for Claude once your domain is verified, and switch off Rate chats under Organization settings > Data and Privacy, because Anthropic may keep a rated chat for up to five years.414 Team accounts arrive by invitation or just-in-time, meaning an account appears when someone first signs in through SSO, so a person still removes the seats of leavers.14
Claude Enterprise audit logs and retention
Owners on Claude Enterprise export 180 days of audit logs from the same settings page, and the export holds identifiers rather than chat contents.15 Custom retention starts at 30 days and queues older chats for permanent deletion on save, so set 90 days unless someone names a reason to keep chats longer.16 A deleted chat cannot leak.
Pick Claude Team while no auditor asks for logs, at $20 per Standard seat a month billed annually, about €18, or $25 billed monthly, about €22.2 Move to Claude Enterprise when a contract or your data protection officer asks for audit logs, SCIM or your own retention period, and budget for at least 20 seats, the Enterprise minimum.2 Enterprise bills every message on top of its $20 seat at API rates, meaning the per-token prices developers pay, so log two months of Team usage before you ask for a quote. Claude Team pricing in euros works through the sums for a team that mixes Standard and Premium seats. A security team weighing OpenAI's offer as well can set these controls against ChatGPT Enterprise in Europe, which offers EU storage for new workspaces and EU inference for eligible customers.
How is Claude safety-trained? Constitutional AI in 60 seconds
Constitutional AI, meaning training a model against a written set of principles rather than only on human ratings, is the method Anthropic published in December 2022.17 The model critiques and rewrites its own answers against those principles, then learns from an AI judge instead of from human labels of harmful outputs.17
Claude's current constitution, published on 22 January 2026, ranks four properties in order: broadly safe, broadly ethical, compliant with Anthropic's guidelines and genuinely helpful.18 Anyone may read and reuse the full text, although Anthropic itself says Claude's outputs do not always match it.18
Anthropic also tests its models on real security work: in two weeks of February 2026, Claude Opus 4.6 found 22 Firefox vulnerabilities, and Mozilla rated 14 of them high severity.19 Turning a flaw into an attack went far worse, since several hundred attempts produced two working exploits, both in a test build with browser protections removed.19 Mozilla's own engineers still triaged and fixed the reports themselves.
What counts as secure enough from a CISO's view?
A chief information security officer, the CISO who signs off new vendors, asks four things about Claude: audited controls, written AI governance, a contract for the data flow and your own team's habits. Anthropic answers the first three with SOC 2 Type II, ISO/IEC 42001 and the DPA, while the fourth stays with you on every plan.
GDPR Article 28(1) lets you use only processors that give sufficient guarantees of appropriate security measures, and the SOC 2 report and ISO certificates are how you document that check.20 Article 32 then asks for security that matches the risk, and SSO, retention and audit logs are the measures inside the workspace that answer it.20
A data protection impact assessment, or DPIA, meaning a written risk analysis before processing starts, is what Article 35 asks for when new technology is likely to put people at high risk.20 The guidelines the European Data Protection Board endorsed say processing that meets two of their nine criteria will in most cases need one, and new technology such as AI is one of the nine.21 Screening Claude against that list takes about two hours, and the written result belongs in the same file as the certificates.
What European companies still have to do themselves
Certificates cover Anthropic's half of the work, and the eight tasks below cover yours, each with a usual owner and our own time estimate. Added up they come to about nine hours across five roles.
| Task | Owner | Our estimate |
|---|---|---|
| 1. Keep company data off personal plans such as Free and Pro | Budget holder | 30 minutes |
| 2. File the DPA, the clauses and the transfer in your Article 30 record | Data protection officer | 1 hour |
| 3. Screen for a DPIA under Article 35 and write the result down | Data protection officer | 2 hours |
4. Verify your domain and tick Require SSO for Claude | IT admin | 1 hour |
5. Switch off Rate chats | IT admin | 5 minutes |
| 6. Set 90-day retention on Enterprise, or a delete rule on Team | IT admin and data protection officer | 30 minutes |
| 7. Write the prompt rule into your AI policy | Leadership team | 2 hours |
| 8. Run a two-hour session for staff on safe prompts | Team leads | 2 hours |
Article 4 of the EU AI Act asks providers and deployers to take measures that support AI literacy among their staff, so task 8 is a measure to record rather than a legal exam.22 The prompt rule in task 7 is the line staff remember, and our AI policy template gives it a home among eight sections.
satori-claude takes tasks 4 to 6 off your desk and runs the workspace for €490 once plus €14 per user a month on top of the licence, €22 Standard or €110 Premium, excl. VAT. Talk to us on a free 30-minute call if you want a second reader on the DPIA screening.
Open trust.anthropic.com, find the row for your plan in the scope table and copy its marks for SOC 2, ISO 27001 and ISO 42001 into your vendor file with today's date. In fifteen minutes you hold a one-page vendor record that answers the first question your auditor will ask.
Sources
Footnotes
-
Anthropic (2026). Anthropic Trust Center: compliance scope by product and resources, checked 4 November 2026. https://trust.anthropic.com ↩ ↩2 ↩3 ↩4 ↩5
-
Anthropic (2026). Plans and Pricing: compare features across plans, checked 4 November 2026. https://claude.com/pricing ↩ ↩2 ↩3 ↩4 ↩5
-
Anthropic (2025). Commercial Terms of Service, effective 17 June 2025. https://www.anthropic.com/legal/commercial-terms ↩ ↩2
-
Anthropic (2026). Is my data used for model training? (commercial products). Anthropic Privacy Center. https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training ↩ ↩2
-
AICPA and CIMA (2026). SOC 2 Reporting on an Examination of Controls at a Service Organization Relevant to Security, Availability, Processing Integrity, Confidentiality, or Privacy. https://www.aicpa-cima.com/cpe-learning/publication/soc-2-reporting-on-an-examination-of-controls-at-a-service-organization-relevant-to-security-availability-processing-integrity-confidentiality-or-privacy ↩
-
Anthropic (2026). What certifications has Anthropic obtained? Anthropic Privacy Center. https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained ↩
-
Anthropic (2025). Anthropic achieves ISO 42001 certification for responsible AI, 13 January 2025. https://www.anthropic.com/news/anthropic-achieves-iso-42001-certification-for-responsible-ai ↩ ↩2
-
Anthropic (2026). Where are your servers located? Do you host your models on EU servers? Anthropic Privacy Center. https://privacy.claude.com/en/articles/7996890-where-are-your-servers-located-do-you-host-your-models-on-eu-servers ↩
-
Anthropic (2026). Data residency. Claude Platform documentation. https://platform.claude.com/docs/en/manage-claude/data-residency ↩
-
Anthropic (2026). Enable US-only inference for your organization. Claude Help Center. https://support.claude.com/en/articles/15422948-enable-us-only-inference-for-your-organization ↩
-
Anthropic (2025). Data Processing Addendum. https://www.anthropic.com/legal/data-processing-addendum ↩
-
Amazon Web Services (2026). Regional availability of models in Amazon Bedrock. https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html ↩
-
Google Cloud (2026). Data residency. Gemini Enterprise Agent Platform documentation. https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency ↩
-
Anthropic (2026). Important considerations before enabling single sign-on (SSO) and JIT/SCIM provisioning. Claude Help Center. https://support.claude.com/en/articles/10276682-important-considerations-before-enabling-single-sign-on-sso-and-jit-scim-provisioning ↩ ↩2
-
Anthropic (2026). Access audit logs. Claude Help Center. https://support.claude.com/en/articles/9970975-access-audit-logs ↩
-
Anthropic (2026). Configure custom data retention controls for Enterprise plans. Anthropic Privacy Center. https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans ↩
-
Anthropic (2022). Constitutional AI: Harmlessness from AI Feedback, 15 December 2022. https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback ↩ ↩2
-
Anthropic (2026). Claude's new constitution, 22 January 2026. https://www.anthropic.com/news/claude-new-constitution ↩ ↩2
-
Anthropic (2026). Partnering with Mozilla to improve Firefox's security, 6 March 2026. https://www.anthropic.com/news/mozilla-firefox-security ↩ ↩2
-
European Union (2016). Regulation (EU) 2016/679 (General Data Protection Regulation), Articles 9, 28, 30, 32 and 35. EUR-Lex. https://eur-lex.europa.eu/eli/reg/2016/679/oj ↩ ↩2 ↩3
-
Article 29 Working Party (2017). Guidelines on Data Protection Impact Assessment (DPIA), WP248 rev.01, endorsed by the EDPB on 25 May 2018. https://ec.europa.eu/newsroom/article29/items/611236/en ↩
-
European Commission (2026). Article 4: AI literacy. AI Act Service Desk. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4 ↩
Frequently asked questions
Claude Enterprise security is the set of audited certificates, contract terms and admin controls that come with Anthropic's business plans. Anthropic's Trust Center lists SOC 2 Type II, ISO 27001 and ISO/IEC 42001 for both Claude Team and Claude Enterprise, and its Data Processing Addendum is part of the Commercial Terms. Enterprise adds SCIM, audit logs, custom data retention, role-based access and IP allowlisting on top of the single sign-on that Team already has.
Ask for the reports instead of trusting a sales deck. Anthropic's Trust Center at trust.anthropic.com offers the 2025 SOC 2 Type 2 report, the 2025 ISO 27001 and ISO 42001 certificates, two 2026 SOC 2 bridge letters and the 2025 penetration testing reports to customers who request access. Your security lead reads the scope, the dates and any exceptions the auditor found, and files the result with your other vendor records.
Claude Team carries the same SOC 2 Type II, ISO 27001 and ISO 42001 coverage as Enterprise, plus single sign-on, domain verification and no training on your content by default. Move to Enterprise when an auditor, a customer contract or your data protection officer asks for audit logs, SCIM provisioning or a retention period you set yourself, because Team has none of the three. Enterprise costs $20 per seat a month on annual billing, about €18, with all usage charged on top at API rates.
No, not by default. Anthropic states that it does not use inputs or outputs from commercial products such as Claude Team, Claude Enterprise and the API to train its models. The exception is feedback, because a thumbs up or down sends that chat to Anthropic, which may keep it for up to five years and train on it. Owners of a Team or Enterprise plan switch member feedback off with the Rate chats setting under Organization settings, then Data and Privacy.
SOC 2 Type II is an auditor's report on whether a company's security controls worked across a whole audit period, so it answers questions about access, availability and confidentiality. ISO/IEC 42001 certifies how a company governs AI itself, meaning a framework for finding, assessing and reducing the risks of its AI systems. A vendor review for an AI tool needs both, because a model can sit on well-run servers and still be badly governed.
The certificates hold for those sectors, while the deciding work sits with you. Health data is special category data under GDPR Article 9, so run a data protection impact assessment under Article 35 before the first patient detail reaches a prompt. Anthropic's HIPAA-ready offering answers US health law and sits on Enterprise only, so it does not settle the European question. A bank or insurer usually needs the audit logs and custom retention of Claude Enterprise, because its own auditors ask who did what and how long it was kept.
About the author
Joachim helps European companies understand and use AI in everyday work. As founder of satori., he trains teams, coaches leadership groups, and shows how tools like Claude become a natural part of work.