Is it illegal to use ChatGPT to process documents with personal data?

Not inherently illegal, but using the free or Plus tier without a DPA to process documents containing personal data of EU residents is a violation of GDPR Article 28. Using ChatGPT Team, Enterprise or API with a signed DPA is compliant, subject to the other checklist items.

Does using Claude for Business make AI document processing GDPR compliant?

It addresses the DPA requirement (Article 28) and the training data concern. You also need to ensure the data transfer mechanism is documented, staff are trained on approved tools, and processing is recorded in your Article 30 register.

What is a Data Processing Agreement and where do I get one?

A DPA is a contract between your organisation (controller) and the AI provider (processor) specifying the terms of personal data processing. For ChatGPT, it is available in the Team and Enterprise plan settings. For Claude, it is available through Anthropic's business agreements. For Microsoft Copilot in M365, it is included in the Microsoft Product Terms.

Can I use Copilot in Microsoft 365 to process personal data under GDPR?

Microsoft 365 commercial subscriptions include a DPA through the Microsoft Product Terms, and Microsoft relies on the EU-US Data Privacy Framework for transfers. Commercial Copilot use is generally compliant from an Article 28 perspective, subject to your organisation's own policies.

How do I reduce GDPR risk when I need to use AI tools with documents?

Use business or enterprise tiers with signed DPAs. Disable model training. Strip unnecessary personal data from documents before uploading — compress and clean files locally using MiniFiles.io. Document the processing in your Article 30 register. For documents where AI is not essential, keep processing local.

GDPR Checklist for Using AI Tools With Personal Documents

The core GDPR requirement: Article 28 and Data Processing Agreements

GDPR Article 28 requires that whenever a controller (your organisation) engages a processor (a third-party service that processes personal data on your behalf), there must be a written contract — a Data Processing Agreement (DPA) — that specifies what data is processed, for what purpose, under what security measures, and with what sub-processors. An AI tool that processes a document containing personal data is acting as a data processor. If you do not have a signed DPA with that AI provider, you are in violation of Article 28 — regardless of whether any actual harm occurs.

Which AI tools offer DPAs — and which do not

OpenAI offers a DPA for ChatGPT Team, ChatGPT Enterprise, and API customers. The free and Plus tiers of ChatGPT do not include a DPA. Anthropic offers a DPA for Claude for Business and Enterprise customers, and for API customers via their usage policies. Claude.ai free and Pro do not include a DPA. Google offers a DPA for Google Workspace customers using Gemini within Workspace. Consumer Gemini (gemini.google.com) does not include a DPA. Microsoft offers a DPA for Microsoft 365 commercial customers using Copilot. Consumer Microsoft accounts do not include a DPA. If you are using the free tier of any of these products to process documents containing personal data of EU data subjects, you are operating without the required legal framework.

What counts as personal data in a document

Under GDPR Article 4, personal data is any information relating to an identified or identifiable natural person. In practice, this includes: names, email addresses, phone numbers, and physical addresses anywhere in a document; employee records, CVs, and HR files; client contracts and invoices with individual names; medical records and health information (which are special category data under Article 9, requiring stricter protection); photographs and images of individuals. If any of these appear in a document you process with an AI tool, GDPR applies.

The training data problem

A separate concern from Article 28 is whether AI providers use your uploaded data to improve their models. Under GDPR, using personal data to train AI models requires a lawful basis — typically consent or legitimate interest, both of which come with significant requirements. OpenAI disables training for Enterprise and API users, and allows opt-out for Free and Plus users. Anthropic disables training for API and Enterprise; consumer accounts may be used for improvement unless opted out. If employees are uploading HR documents or client files to consumer AI products without opting out of training, the organisation may be enabling personal data to be processed for a purpose (model training) for which they have no legal basis.

International data transfers

OpenAI, Anthropic, and Google process data on US-based infrastructure. Transferring personal data of EU residents to the US requires either: an adequacy decision (the EU-US Data Privacy Framework covers US organisations that have self-certified, including Google and Microsoft), Standard Contractual Clauses (SCCs), or binding corporate rules. OpenAI relies on SCCs for EU-US transfers. Anthropic relies on SCCs. Google and Microsoft rely on the EU-US Data Privacy Framework. If your DPA does not address transfer mechanisms, or if you are using a tier without a DPA, the transfer may lack adequate safeguards — the same category of violation that produced Meta's €1.2 billion fine.

GDPR compliance checklist for AI document workflows

Before using any AI tool to process documents containing personal data, work through this checklist. (1) Does the AI provider offer a DPA for your tier? If not, upgrade or switch tools. (2) Have you signed the DPA? Some providers make this self-service; others require a request. (3) Does the DPA address international data transfers with appropriate SCCs or framework references? (4) Have you documented the processing activity in your Records of Processing Activities (Article 30)? (5) Have you disabled model training, or confirmed it is disabled on your tier? (6) Have you informed staff of which AI tools are approved for processing personal data, and which are not? (7) For special category data (health, biometric, racial or ethnic origin), do you have explicit consent or another Article 9 basis?

When to keep files local instead of using AI

For documents where AI analysis is not actually necessary, the simplest compliance approach is not to upload them at all. Many workflows that involve AI tools can be separated: use AI to draft templates, structure outputs, or summarise anonymised data — and keep the personal data processing local. For compression, format conversion, password protection, and size reduction of sensitive documents, browser-based local tools like MiniFiles.io process files on the device without any data transfer. The document never reaches an external server, which means no Article 28 requirements are triggered.