Is It Safe to Upload Sensitive Documents to AI Tools? A UK GDPR Guide

AI tools have become a routine part of how many businesses work. Documents get summarised, contracts get reviewed, emails get drafted — often with personal data included, often without much thought about where that data goes.

This guide sets out the specific UK GDPR risks of uploading sensitive documents to AI tools, what questions you should be asking before you do, and what a compliant approach looks like in practice.

What's the actual risk?

When you upload a document to an AI tool — whether that's ChatGPT, Claude, Google Gemini, Microsoft Copilot, or any other service — several things may happen with the data in that document:

• It may be transmitted to and processed on servers located outside the UK or EU
• It may be used to train or improve the AI model, depending on the service's terms
• It may be retained by the service provider for a period after your session
• It may be accessible to employees of the provider for safety, quality, or legal review purposes

For documents that contain personal data — employee records, client contracts, financial information, medical details, legal correspondence — each of these points creates a potential UK GDPR issue.

Is using an AI tool "processing" under UK GDPR?

Yes. Under UK GDPR, "processing" means any operation performed on personal data — including storing, transmitting, using, or disclosing it. When you upload a document containing personal data to an AI service, you are processing that data and causing it to be processed by a third party.

This triggers several obligations:

1. You need a lawful basis. Under Article 6, you need a lawful basis for every processing activity. Uploading an employee's performance review to an external AI tool is unlikely to be covered by your existing lawful basis for holding that data.

2. The AI provider becomes a data processor. Under Article 28, any third party that processes personal data on your behalf is a data processor. You are required to have a Data Processing Agreement (DPA) in place with them before sharing personal data.

3. International transfers may apply. If the AI service processes data on servers outside the UK or EU/EEA, this constitutes a restricted international transfer requiring a transfer mechanism in place.

4. Your privacy notice may not cover it. If your privacy policy tells individuals their data is used for X, and you're now feeding it to an AI tool, individuals may not have been informed of that use.

The enterprise vs. consumer AI distinction

Consumer or free-tier AI tools (e.g. free ChatGPT, standard web-based Claude) are designed for individual use. Their terms typically allow the provider to use your inputs to improve the model, may not include a DPA, and may store your conversations indefinitely. These tools are generally not appropriate for processing personal data about third parties.

Enterprise AI tools (e.g. Microsoft 365 Copilot, ChatGPT Enterprise, Google Workspace AI) typically include a DPA and commit not to use your data for model training. They are more likely to be compliant — but you still need to verify the specific terms, check for international transfer provisions, and assess whether the processing is covered by your lawful basis.

The tool itself is not the only question. The terms of service, the data processing agreement, and the transfer mechanisms are what determine compliance.

What to check before uploading personal data to any AI tool

Practical scenarios

Scenario 1: Summarising a supplier contract

A contract between your business and a supplier, containing company details and supplier contact names and payment terms.

Risk level: Lower. Business contact information of individuals acting in a professional capacity is generally lower-risk. Check your DPA situation, but this is a manageable use case.

Scenario 2: Reviewing an employee's performance appraisal

Contains the employee's name, role, salary, and performance commentary.

Risk level: Higher. Employment data is sensitive. The employee almost certainly hasn't been told their appraisal will be shared with an AI provider. This is unlikely to be covered by any existing lawful basis without specific steps being taken.

Scenario 3: Analysing customer complaint emails

Contains customer names, contact details, and descriptions of their issues.

Risk level: Higher. Customers have not consented to their communications being processed by a third-party AI. A lawful basis assessment and updated privacy notice would be needed before doing this at scale.

Scenario 4: Processing medical or health documents

Any document containing health, medical, or disability information.

Risk level: High. Health data is "special category" data under Article 9, subject to stricter rules and requiring an explicit lawful basis. Uploading this to a general-purpose AI tool without careful analysis is a significant compliance risk.

A compliant approach: process first, share later

The most robust way to use AI tools with sensitive documents is to ensure that personal data is identified and handled before it reaches any external service. This means scanning documents for PII before processing, redacting or pseudonymising personal data where the AI task doesn't require it, using AI tools with compliant DPAs and data residency commitments, and keeping an audit trail of what was processed and under what basis.

Frequently asked questions