GDPR-compliant AI for companies

No AI tool makes a company GDPR-compliant on its own. Compliance depends on how the company uses the tool: which data goes in, on what legal basis, and under which contract. A provider can supply the technical and contractual groundwork. This page separates what Langdock provides from what remains your responsibility as the controller.

Written by LangdockUpdated

  • This page is general information, not legal advice. Have your own use cases assessed by your data protection officer or legal counsel.

What does the GDPR require when a company uses AI?

The GDPR applies as soon as personal data is processed, for example when employees enter customer names, e-mails or application documents into a chat. The company that decides on this use is the controller; the AI provider usually acts as a processor. The main requirements are:

GDPR requirements for the use of AI tools
RequirementWhat it means
Legal basis (Art. 6)Every processing of personal data needs a legal basis, such as a contract, a legitimate interest or consent.[1]
Processor contract (Art. 28)A provider that processes personal data on your behalf needs a data processing agreement and may only act on your instructions.[1]
Transfers outside the EU (Art. 44 ff.)A transfer to a country outside the EU/EEA is only allowed if the conditions of Chapter V are met: an adequacy decision (Art. 45), appropriate safeguards such as standard contractual clauses (Art. 46) or, in specific situations, a derogation (Art. 49).[1]
Security of processing (Art. 32)Technical and organisational measures appropriate to the risk, such as encryption and access control.[1]
Records and transparency (Art. 13, 14, 30)As a rule, the use belongs in the record of processing activities; Art. 30(5) exempts organisations with fewer than 250 employees unless, among other things, the processing is not occasional. Data subjects must be informed about the processing unless an exception in Art. 13(4) or 14(5) applies, for example because they already have the information.[1]
Data protection impact assessment (Art. 35)Required before processing that is likely to result in a high risk for the people concerned.[1]

What does Langdock provide for GDPR compliance?

What Langdock provides, by requirement
RequirementWhat Langdock provides
Processor contractA published data processing agreement (DPA).[3]
Hosting locationThe standard cloud version is hosted via Microsoft Azure on servers inside the EU. The application and most models are hosted entirely in the EU.[2]
Model trainingCustomer data is never used to train or improve AI models.[2]
Security of processingEncryption with AES-256 at rest and TLS 1.2+ in transit; ISO 27001 certification and a SOC 2 Type II audit.[2]
Access controlSSO, SAML and SCIM in every plan, so access follows your identity provider.[4][5]
Deployment choiceBesides the EU cloud, as Enterprise options: a dedicated deployment managed by Langdock, a deployment in your own cloud, or on-premise.[2]
Your own model contractsWith Bring Your Own Keys (BYOK) the workspace uses only the models you configure, under your own contract with the model provider.[6]

Which models have no EU-hosted version?

Not every model has an EU-hosted version. 41 of the 47 models currently listed have a version hosted in the EU. For 6 models the model list shows no EU-hosted version, listed here with the region it shows instead: DeepSeek v3.1 (Global), Gemini 3.1 Pro Preview (Global), Gemini 3 Flash Preview (Global), GPT-5 Pro (Global), GPT oss (120b) (Global), and Llama 4 Maverick (Global). The model list shows the region of each model and is the current reference.

If prompts may contain personal data, check the region of a model before you make it available, and decide whether a model without an EU-hosted version fits your transfer rules. EU hosting is not a GDPR requirement in itself: transfers outside the EU are lawful when the conditions of Chapter V are met.

What remains your responsibility?

As the controller you decide which data is processed and why, and the responsibility for these tasks stays with you. A processor has to assist you with some of them, for example with data subject requests and impact assessments (Art. 28(3)(e) and (f)), but it cannot take over your responsibility:

  • Define the legal basis for each use case in which personal data is processed.
  • Conclude the data processing agreement and file it with your processor documentation.
  • Add the use to your record of processing activities, if you have to keep one, and update your privacy notices for employees and customers.
  • Check whether a data protection impact assessment is needed, for example for HR data or other sensitive use cases.
  • Set internal rules: which data may be entered, which models are enabled, who may build and share agents and integrations.
  • Involve the right people early: your data protection officer and, where one exists, employee representatives.
  • Handle data subject requests such as access and erasure for the data you put into the tool.
  • Review the setup regularly, especially when new models, integrations or use cases are added.

What should you ask any AI provider?

  • Is there a data processing agreement, and which sub-processors does it name?
  • Where are the application and each model hosted, and can models outside the EU be switched off?
  • Is customer data used to train or improve models?
  • Which certifications and audit reports exist, and can you read them?
  • How are users provisioned and removed (SSO, SCIM), and which admin controls exist?
  • Which deployment options exist if the standard cloud does not fit your requirements?

Langdock’s answers to these questions are on the facts page and the security page.

Frequently asked questions

No tool is GDPR-compliant on its own. Langdock offers a data processing agreement under Art. 28 and security measures such as encryption (Art. 32). On top of what the law requires, it hosts the standard cloud version on servers inside the EU and does not use customer data for model training. Whether your use of Langdock is compliant depends on your own legal basis, documentation and internal rules.

41 of the 47 models currently listed have a version hosted in the EU. For 6 models the model list shows no EU-hosted version, listed here with the region it shows instead: DeepSeek v3.1 (Global), Gemini 3.1 Pro Preview (Global), Gemini 3 Flash Preview (Global), GPT-5 Pro (Global), GPT oss (120b) (Global), and Llama 4 Maverick (Global).

No. Customer data is never used to train or improve AI models.

Yes. Whenever a provider processes personal data on your behalf, Art. 28 GDPR requires a data processing agreement. Langdock publishes its agreement on the DPA page.

Sources

  1. EUR-Lex: Regulation (EU) 2016/679 (General Data Protection Regulation) (checked on )
  2. Langdock: Security
  3. Langdock: Data processing agreement (DPA)
  4. Langdock: Pricing
  5. Langdock documentation: SAML SSO with Microsoft Entra
  6. Langdock documentation: Bring Your Own Keys (BYOK)