Skip to content
Outrich Data

The GDPR and selling company data to AI labs

Updated 2026-11-16 · 7 min read · By the Outrich Data team

The GDPR does not ban licensing company data for AI training, but any personal data in it needs a lawful basis, a compatible purpose, transparency and a valid transfer route. Truly anonymous data is outside the GDPR. Pseudonymized data is not. A processor cannot license its clients' data. This is general information, not legal advice.

This page is general information, not legal advice. Whether your company can license a given dataset depends on your contracts, your privacy notices and the law where you and your customers are. Review it with your own counsel before signing anything.

When the GDPR applies to a seller

The GDPR applies to organizations established in the European Union when they process personal data. It also reaches organizations outside the EU that offer goods or services to people in the EU or monitor their behavior. The United Kingdom keeps its own version, the UK GDPR, which follows the same structure with some differences that have grown since Brexit.

A company outside Europe can still hold data within scope. A US firm with a Dublin office, EU employees or EU customers will usually have records the GDPR covers. Those records need separate handling in any data sale.

This guide is general information, not legal advice. It describes the main questions in plain terms. National laws, regulator guidance and court rulings add detail that a summary cannot carry. Take advice from qualified counsel in the relevant country.

Personal data and anonymous data

Personal data is any information relating to an identified or identifiable living person. Names and email addresses are obvious cases. So are staff IDs, voice recordings, and free text that describes a person well enough to single them out.

Recital 26 marks the boundary. The principles of data protection do not apply to anonymous information, meaning information that does not relate to an identifiable person, or personal data made anonymous so that the person can no longer be identified. To judge identifiability, account is taken of all the means reasonably likely to be used, by the controller or by anyone else, considering cost, time and available technology.

The practical point is that anonymous data can be licensed without the GDPR applying to the dataset. The bar is high, and regulators read it strictly. Regulators have also taken the view that the act of anonymizing personal data is itself processing that needs a lawful basis.

Three risks are commonly used to test a claim of anonymity. Singling out means isolating one person's records in the dataset. Linkability means connecting records about the same person, within the dataset or across datasets. Inference means deducing something about a person with significant probability. A dataset that resists all three is a strong candidate. One that fails any of them is probably not anonymous.

Pseudonymization is a safeguard, not an exit

Pseudonymization means processing personal data so it can no longer be attributed to a person without additional information that is kept separately and protected. Replacing names with codes and hashing email addresses are common examples.

Recital 26 says that pseudonymized data which could be attributed to a person by the use of additional information should be treated as information on an identifiable person. For the seller who holds the key, it remains personal data. Whether it is personal data in the hands of a buyer with no access to the key is a developing question in European case law. A cautious seller treats it as personal data unless counsel advises otherwise.

Controller or processor

The controller decides why and how personal data is processed. A processor handles it on the controller's behalf and only on the controller's documented instructions. The role is decided by the facts for each set of data, and a company is often both.

A processor that starts deciding purposes for itself is treated as a controller for that processing and is in breach of its processing agreement. For a service business, this is usually the end of the matter for client data.

DataTypical roleCan the company license it
Its own employees' work recordsControllerPossibly, if the other GDPR conditions are met
Its own customers' account and support recordsControllerPossibly, if the other GDPR conditions are met
Client data handled to deliver a serviceProcessorNo, not without the client's decision and instruction
Data made truly anonymousOutside the GDPR once anonymousYes, subject to contracts and other law

Lawful basis, purpose limitation and compatibility

Article 6 requires a lawful basis for each processing purpose. Supplying personal data to a third party for AI training is a new purpose. Consent is one basis, though it is hard to collect after the fact and hard to rely on for employees. Legitimate interests is another. It needs a documented balancing test, and European regulators have said it can support some AI training but not every case.

Article 5 adds purpose limitation. Data collected for one purpose must not be further processed in a way incompatible with it. Article 6(4) lists the factors for judging compatibility.

Special categories of data, such as health data, trade union membership and data revealing ethnic origin, face tighter rules under Article 9 and are best excluded.

Employee data deserves a specific caution. Because of the imbalance between employer and worker, European regulators treat employee consent as rarely freely given. Several member states also give works councils a say in how staff data and monitoring systems are used. A plan that depends on staff consent is fragile.

  • Any link between the original purpose and the new one.
  • The context of collection and the relationship between the people and the controller.
  • The nature of the data, especially special categories.
  • The possible consequences for the people concerned.
  • The existence of safeguards, which may include encryption or pseudonymization.

Transparency and the right to object

People must be told what is done with their data. Privacy notices have to state the purposes, the legal basis and the recipients or categories of recipients. Where a controller intends to process data for a new purpose, it must inform the people concerned before it does so.

Where the basis is legitimate interests, individuals have the right to object, and the controller must then stop unless it can show compelling grounds. A seller needs a working way to take objections and to remove those people's records before delivery. A notice changed quietly on the eve of a sale is unlikely to satisfy a regulator.

Fines under the GDPR can be large, but for most sellers the nearer risks are practical: a regulator's order to stop, complaints from staff or customers, and a buyer that walks away when it cannot document a lawful source. AI developers serving the European market face their own documentation duties and increasingly ask suppliers to show the legal basis for the data.

International transfers

Most AI labs are outside Europe. Sending personal data to a country outside the European Economic Area requires a transfer mechanism under Chapter V of the GDPR. The main routes are an adequacy decision for the destination country, or appropriate safeguards such as the European Commission's standard contractual clauses supported by a transfer risk assessment.

For the United States, the Commission adopted an adequacy decision in 2023 that covers organizations certified under the EU-US Data Privacy Framework. Its durability has been challenged in court, so check its current status and whether the buyer is certified. The UK runs its own transfer rules and its own contract terms. Anonymous data needs no transfer mechanism, which is one more reason sellers aim for it.

The seller remains responsible for the transfer it makes. It is not enough that the buyer says it complies.

DPIAs and what to settle before a deal

A data protection impact assessment is required where processing is likely to result in a high risk to individuals. Large-scale reuse of staff or customer records for AI training may meet that test, and many regulators would expect one. The assessment describes the processing, tests necessity and proportionality, and records the risks and the measures taken.

A UK or EU seller should have written answers to the following before signing. Outrich Data asks for these during its data title search, and a seller's counsel should own the conclusions.

  • For each dataset, whether the company is controller or processor.
  • Whether the delivered data is anonymous, and the evidence for that.
  • If personal data remains, the lawful basis and the compatibility analysis.
  • What the privacy notices said when the data was collected, and what has been communicated since.
  • How objections and deletion requests are honored before and after delivery.
  • The transfer mechanism, if personal data leaves the UK or the EEA.
  • Whether a DPIA was done, and whether works councils or employee representatives must be consulted.

Common questions

Does the GDPR ban selling data for AI training?
No. It sets conditions. Personal data needs a lawful basis, a compatible purpose, transparency and a valid transfer route. Truly anonymous data is outside the GDPR. Many planned deals fail the conditions for some of the data and pass for the rest.
Is hashed or coded data anonymous under the GDPR?
Generally no. That is pseudonymization. If someone holds the additional information needed to link the codes back to people, Recital 26 treats the data as relating to identifiable persons. It lowers risk but the GDPR still applies.
Can a processor license client data if it is de-identified first?
Not on its own decision. A processor acts on the controller's instructions, and de-identifying data for the processor's own commercial use is outside them. The client would need to agree, and the client then needs its own lawful basis.
Is the UK position the same as the EU's?
Largely, but not exactly. The UK GDPR has the same core principles and its own regulator, transfer tools and recent legislative changes. A company with both UK and EU data should have each reviewed under its own regime.

Find out if your data qualifies.

Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.

esc