Is it legal to sell company data to AI companies?

Selling company data to AI companies is usually legal when three things line up: you own or control the records, your contracts and terms allow it, and personal data inside them is handled lawfully. This guide sets out each question. It is general information, not legal advice.

  • Guide
  • By
    Keilian Knudsen
  • 7
    min read
  • Updated
    6 Nov 2026
A bright white rectangular frame of points around a faint grid on black

Key takeaways

  • Under GDPR Article 28, a processor may only handle personal data on the controller's documented instructions, so client data you hold as a supplier is usually not yours to license.
  • Slack's API terms, effective 10 October 2025, bar apps offered outside your organisation from using API data to train a large language model.
  • The FTC warned in February 2024 that quietly changing terms to allow AI training may be unfair or deceptive.
  • Data protection law does not apply to truly anonymous information, but pseudonymised or de-identified data can still be personal data.

The three questions to settle first

No single law says yes or no. Whether a deal is lawful depends on the records, the contracts around them and the people in them. Most legal reviews come down to three questions, and a no to any one of them can stop a deal or shrink it.

The table sets out the three questions, what to check and who usually makes the call. It assumes you already know what a data licence is and how deals are structured. This page is general information, not legal advice, so use it to prepare for a conversation with your lawyer.

The three questions to settle first
QuestionWhat to checkWho usually decides
1. Do you own or control the records?Whether each record is company work, an employee's personal material or a client's data you hold for themFounders or directors, with legal counsel
2. What do your contracts and terms say?Customer contracts, platform and supplier terms, employment contracts, NDAs and your privacy noticesLegal counsel, sometimes the customer or platform
3. How is personal data handled?Lawful basis, purpose, special categories, de-identification and sale rules such as the CCPAYour data protection lead or DPO, with counsel

Do you own or control the records?

Start by sorting what you hold into three piles. Company records are made by staff doing their jobs. In the US, the Copyright Office says work prepared by an employee within the scope of their employment is a work made for hire, owned by the employer. In the UK, section 11 of the Copyright, Designs and Patents Act 1988 gives the employer first ownership of work made in the course of employment, unless agreed otherwise.

Employee data is different. Ownership of a document does not settle the privacy of the people named in it, such as HR files, health notes or private messages in a work tool. Those are covered by question three.

Staff privacy is also where public pushback starts. When Spirit Airlines' internal emails and Teams messages were put up for sale in its bankruptcy, 121 members of Congress urged Spirit and Google to protect former employees' privacy, the Las Vegas Sun reported. Their letter warned that removing names, email addresses or other direct identifiers does not necessarily make a dataset anonymous. A record can belong to the company and still be sensitive for the people in it.

Client data is the pile that most often stops a deal. If you hold personal data for a client as a processor, GDPR Article 28 says you may process it only on the controller's documented instructions. A processor that decides its own purposes for the data is treated as a controller for that processing under Article 28(10), and takes on the controller's duties. Agencies, BPOs and software vendors should also check who owns deliverables and client files under each contract.

What do your contracts and terms say about AI training?

Even records you own can be locked by what you have signed. Read each agreement for confidentiality, data use, AI and machine learning clauses before any sample leaves the building.

Contracts tend to bite in two ways. Some forbid the use outright, such as training AI on data received under them. Others limit the route, for example by banning bulk export through an API, so the same records may be licensable only if you export them another way or ask permission first.

  • Customer contracts. Many limit customer information to delivering the service. Data processing agreements usually go further and restrict any other use.
  • Platform and supplier terms. Slack's API terms, effective 10 October 2025, say a provider of an app offered outside its organisation may not use API data to train a large language model or bulk export message and file data without a further agreement. Shopify's updated API terms, effective 27 February 2026, say merchant and customer data may not be used to train AI without written consent from Shopify or the merchant.
  • Employment contracts. Check the IP and confidentiality clauses, and whether staff were told their work messages could be used this way.
  • NDAs. Confidential information received from partners, suppliers or prospects usually cannot be passed on.
  • Privacy notices. Your notice tells people what you do with their data. Licensing to an AI company is often a use it never mentioned.

Can you change your terms to allow AI training?

Only with care. In February 2024 the US Federal Trade Commission warned that it may be unfair or deceptive for a company to adopt more permissive data practices, including using data for AI training, and tell people only through a surreptitious, retroactive change to its terms or privacy policy.

The FTC pointed to its 2023 case against 1Health, which it said broke the law by retroactively widening who it could share personal data with. A new term can cover data collected after people agree to it. Records gathered under the old promise usually need a separate answer, such as consent or removing the personal data.

How must personal data be handled before licensing?

Most company records name people: staff, customers, suppliers. Under GDPR Article 5, personal data must be collected for specified purposes and not further processed in a way that is incompatible with them. Article 6(4) lists what decides whether a new use is compatible, starting with the link between the old and new purposes and the context the data was collected in. If licensing fails that test, you need a fresh lawful basis.

Special categories need more. Article 9 prohibits processing data revealing health, ethnic origin, trade union membership, religious beliefs, sex life or sexual orientation, plus genetic and biometric data, unless a listed exception applies. Most sellers simply take these out.

De-identifying is not the same as anonymising. The UK ICO says data protection law does not apply to anonymous information, but pseudonymisation, such as swapping names for reference numbers, is similar to what other contexts call de-identification, and the result can still be personal data. Ask what a buyer's "de-identified" means in practice.

In California, the CCPA, as amended by the CPRA, defines a "sale" as making a consumer's personal information available to a third party for money or other valuable consideration. The California Privacy Protection Agency says the law covers employees, job applicants and business contacts as well as customers. Information counts as deidentified only if the business takes reasonable measures, publicly commits not to reidentify it and binds recipients by contract to the same rules.

Where do the country and sector rules sit?

Privacy laws, regulators and transfer rules differ by country, and most buyers are American, so sending data abroad is its own step. The table below lists the main rules in each country.

Where do the country and sector rules sit?
Where you are basedMain rules
United StatesCCPA as amended by the CPRA, other state privacy laws, HIPAA, GLBA, FERPA and COPPA
United KingdomUK GDPR, the Data Protection Act 2018, the Data (Use and Access) Act 2025 and the UK-US data bridge
European UnionGDPR Articles 5, 6 and 9, Chapter V transfers and the EU AI Act
CanadaPIPEDA, Quebec's Law 25 and the Alberta and British Columbia acts
AustraliaThe Privacy Act 1988, the Australian Privacy Principles and cross-border disclosure
United Arab EmiratesThe federal PDPL, plus the DIFC and ADGM data protection laws

Which industries have extra rules?

Some sectors carry duties on top of general privacy law. Healthcare companies face HIPAA in the US, where selling identifiable health information needs patient authorisation. Law firms and legal teams must protect privilege and client confidentiality, which only the client can waive. Recruiters and staffing firms hold candidate data, including protected characteristics, that usually stays out. Each industry page lists what typically has to be removed.

What should you ask your lawyer before licensing data?

Take these questions to a lawyer who knows data protection in your country. Clear answers make any conversation with a buyer faster, and they show which records can go into a deal. Write the answers down: a buyer's own lawyers will usually ask the same questions during due diligence.

  • Which records do we own outright, and which do we hold for clients as a processor?
  • Do any customer, supplier, platform or partner agreements limit AI training, export or onward sharing?
  • Do our employment contracts and staff privacy notices cover this use, or do we need to tell staff or ask them?
  • Is licensing compatible with why we collected the personal data, and if not, what lawful basis applies?
  • Which special category or sensitive data must come out, and who checks that it has?
  • Will the buyer get anonymous or only pseudonymised data, and does a CCPA "sale" or an international transfer apply?
  • What warranties and indemnities on rights and privacy will the licence ask us to give?

Check whether your records could qualify

Once you know which records are clearly yours to license, the next question is whether buyers want them. Selling company data to AI companies starts with that list. Pangea.ai's free data check asks about your tools, team size, history and where you are based.

Frequently asked questions

Not in itself, but it is tightly controlled. In the EU and UK you need a lawful basis, a purpose compatible with why the data was collected, and extra conditions for special categories such as health data. In California the CCPA lets people opt out of the sale of their personal information. Selling data in breach of these rules, or of a promise in your privacy notice, can be unlawful.

Yes, within limits. Privacy law still protects the people named in the records. In the EU and UK the company needs a lawful basis and must be open about who receives the data, and California residents can opt out of the sale of their personal information under the CCPA. So the personal data inside the records has to be handled lawfully before anything is shared.

Generally yes, if the processing itself is lawful. Using an AI tool to work on your own data is different from licensing that data so an AI company can train its models. Training is a new purpose and usually involves a new recipient. That is why buyers and their lawyers look closely at your lawful basis, your privacy notices and whether the data is anonymous or only pseudonymised.

It can license records it controls, subject to its contracts and privacy duties. The messages are usually company records, but they name staff, customers and partners, and may contain client information covered by NDAs. Check the platform's own terms for how data may be exported, decide which channels and mailboxes stay out, and take advice on staff privacy before sharing a sample.

Not always, but staff privacy cannot be ignored. Work messages are usually company records, yet they contain personal data about the people who wrote them. Under GDPR, consent is rarely a reliable basis for staff because of the power imbalance, so most companies rely on another basis, update their notices or remove identifying details. Rules differ by country, so check with a local lawyer.

Sources
14

Keep reading

Abstract white lattice of lines and points on black, with a bright square at the centre

What is AI data licensing?

Data licensing gives an AI company the right to use a copy of your records for a set purpose. See how it differs from selling and how deals are structured.

Two framed white grids on black joined by one bright line of points

Data licensing agreements: what they should cover

What goes in a data licensing agreement for AI training: scope, exclusivity, deletion, fees, warranties and liability. Use our checklist before you sign.

A white grid of points on black that breaks up into scattered points

Selling a shut-down company's data to AI labs

Closing or already closed? A shut-down startup's Slack, email and project history can still have value. See what founders should check first.