Get a free valuation

020 3432 4554support@sleepingasset.com

Glossary

AI data licensing glossary

The short answer

What is AI data licensing? It is granting an AI developer, or a data company that supplies one, the right to use a defined copy of an organisation's records to train or test AI models, on written terms, while the organisation keeps ownership.

This glossary defines the terms used when UK companies and insolvency practitioners license business records this way, from anonymisation and pseudonymisation to RL environments and SIP 9. Each definition stands on its own, cites its source and links to the Sleeping Asset guide that goes deeper. Nothing here is legal advice.

A

Acceptance window

The period after delivery in which a buyer checks licensed data against what was agreed, before it has to pay. Terms vary: micro1, for example, says it has 30 days to review and approve transferred data, then pays within 30 days of the seller's invoice1. Ask for agreed acceptance tests, and for silence at the end of the window to count as acceptance.

More: licence terms.

Administration

A UK insolvency procedure in which an insolvency practitioner, the administrator, takes control of a company. The administrator must aim to rescue the company as a going concern, otherwise to get creditors a better result than an immediate winding up, or failing that to realise property for secured or preferential creditors3. Administrators may sell or otherwise dispose of company property40.

More: office-holder guide.

Agent trajectory

A recorded run of an AI agent working through a task: each step, the tools it used and the outcome. Buyers treat trajectories as training data. micro1 says every agent task "produces a trajectory which is valuable training data for showing how real work gets done", and offers to license qualifying trajectories from a company's own AI agents for USD 1 million to 2 million a year1.

More: what AI companies buy.

AI data licensing

Granting an AI developer, or a data company that supplies developers, the right to use a defined copy of an organisation's records to train or test AI models, on written terms, while the organisation keeps ownership. Buyers in this market generally license rather than buy: micro1 says companies "retain ownership of their underlying data"1. Sleeping Asset is a UK-based licensing agent acting for sellers in these deals.

More: our service and what company data is worth.

Anonymisation

Processing personal data so that the risk of anyone being identified from it is sufficiently remote. Effectively anonymised information falls outside UK data protection law, but the ICO says applying anonymisation techniques still "counts as processing personal data", so the step needs a lawful basis18. The test is practical: which means are reasonably likely to be used, and by whom, to identify someone19.

More: anonymisation standard.

B

Bona vacantia

Latin for "ownerless goods". When a UK company is dissolved, the property and rights it still held, other than property held on trust for others, are deemed bona vacantia and belong to the Crown or, as the case may be, the Duchy of Lancaster or the Duke of Cornwall11. Anything of value, including the company's rights in its records, should be dealt with before dissolution.

More: closing your company.

C

Compulsory liquidation

A winding up ordered by the court, for example because the company cannot pay its debts4. In England and Wales the official receiver becomes liquidator when the order is made, and stays in office until another liquidator is appointed5. From then on the liquidator, not the directors, decides whether the company's records are realised, using the power to sell any of its property41.

More: office-holder guide.

Controller (data controller)

The person or organisation that, alone or jointly with others, determines the purposes and means of processing personal data12. A company is normally the controller of the personal data in its own records, so it carries the legal duties when those records are prepared for licensing. On insolvency, the ICO's data sharing code describes an insolvency practitioner "taking the role of controller for the time being"39.

More: UK law guide.

Creditors' voluntary liquidation

A voluntary winding up in which the directors have not made a statutory declaration that the company can pay its debts in full7. It is the procedure used most often: of 22,455 insolvencies in England and Wales in 2025 (business insolvencies, a different count from the 23,942 registered company insolvencies43), 17,530 were creditors' voluntary liquidations35. The liquidator realises the company's assets for creditors, which can include licensing a de-identified copy of its records.

More: for insolvency practitioners.

D

Data broker

In UK usage, usually a business that gathers or buys personal information about individuals and trades it, often for marketing. In 2020 the Information Commissioner called data broking "a complex ecosystem where information appears to be traded widely, without consideration for transparency"37. Sleeping Asset does not trade personal data: acting for the seller, it helps companies license business records with personal data removed.

More: choosing a licensing agent.

Data lab

A term used, including on this site, for the data companies that buy, prepare and supply training data to AI developers, as distinct from the AI labs that build models. Most buyers of company records license onward: micro1 says a company's data is licensed "exclusively to labs and enterprises based in the U.S. and allied countries"1. Ask who ultimately receives your data.

More: buyers compared.

Data licensing agreement

The contract under which a seller grants a buyer rights over a defined copy of its records, setting scope, permitted uses, exclusivity, term, deletion, privacy promises, acceptance, price and liability. Under UK copyright law a licence binds later owners of the copyright, except a good-faith purchaser without notice, while an outright assignment must be in writing and signed23. Also spelled data license agreement.

More: licence checklist.

Data manifest

A short inventory a company gives a prospective buyer before any data moves, listing its systems, the kinds of records held, how many years they cover and roughly how much there is. Cognor Labs describes its manifest as a 10 to 15 minute form31, and micro1 asks a referred company to complete one when it registers2. A manifest holds metadata only, never the records themselves.

More: how it works.

Data partnership

The name several buyers give their programmes for licensing records from companies, such as micro1's Enterprise and Defunct Company Data Partnerships2 and Scale AI's Data Partnerships, for which Scale prefers 40 or more employees30. Despite the word, it is normally a licence on the buyer's paper, not a joint venture, and deserves the same scrutiny as any contract.

More: buyers compared.

Data (Use and Access) Act 2025

The UK Act that amended UK GDPR and the Data Protection Act 2018. It received Royal Assent on 19 June 2025, most of its data protection changes took effect on 5 February 2026, and from 30 September 2026 it replaced the office of the Information Commissioner with the Information Commission17. It also added a lawful basis for a closed list of recognised legitimate interests15.

More: UK law guide.

De-identification

A general term for removing or masking information that identifies people. UK data protection law uses it only in the offence of re-identifying "de-identified personal data"21, and the ICO does not encourage it as a synonym for anonymous or pseudonymous data18. Sleeping Asset uses it for the work of removing personal data, and calls a dataset anonymised only when the legal test is met.

More: anonymisation standard.

In civil litigation in England and Wales, a party discloses a document by stating that it exists or has existed28; US lawyers call this discovery. A party receiving disclosed documents may use them only for those proceedings, unless they were referred to at a public hearing, the court permits, or the disclosing party and the owner agree28. So they are not a law firm's to license.

More: law firms.

E

Exclusivity

A licence term that stops the seller licensing the same records to anyone else, for a period or within a field. UK copyright law defines an exclusive licence as one in writing, signed by the owner, excluding everyone including the owner24. Labelbox Prism says most of its partnerships are exclusive by default32; Cognor Labs says exclusivity typically lasts 24 months31. Limit it and get paid for it.

More: licence terms.

F

Field of use

The uses a licence permits, such as training, fine-tuning or evaluating AI models, and by implication everything it does not. A narrow field protects the seller and can leave room for other licences of the same records. Endless Data's terms for its own customers, for example, tie each licence to a stated "Permitted Use" and treat model training as a separate permission that the order must designate33.

More: licence terms.

I

Information Commission (ICO)

The UK's data protection regulator. On 30 September 2026 the office of the Information Commissioner was abolished and its functions transferred to the new Information Commission, under commencement regulations made under the Data (Use and Access) Act 202517. It still uses the initials ICO, and its anonymisation and data sharing guidance remains the main reference for sellers and office-holders18,39.

More: UK law guide.

Insolvency practitioner

A person qualified under the Insolvency Act 1986 to act as a company's liquidator, provisional liquidator, administrator, administrative receiver or monitor, or as nominee or supervisor of a voluntary arrangement9. Acting as one without being qualified is a criminal offence42. Once appointed to a case, the insolvency practitioner is the office-holder who decides whether the company's records are realised, and on what terms.

More: for insolvency practitioners.

L

Legitimate interests

A lawful basis under UK GDPR for processing personal data that is necessary for the legitimate interests of the controller or a third party, unless overridden by the interests, rights and freedoms of the people concerned, especially children13. It is the usual candidate basis for de-identifying staff and customer data before licensing, and relying on it means weighing those interests and recording the result.

More: UK law guide.

Licence and sale

Two ways of dealing in rights over records. A sale, or assignment, transfers ownership and, for copyright, must be in writing signed by or for the seller. A licence grants permission to use on stated terms while the owner keeps the rights, and binds later owners except a good-faith purchaser without notice23. Licensing is the normal structure for company records. Licence is the noun; license the verb.

More: UK law guide.

Linkability

The ability to combine records about the same person, in one system or across several, so that details harmless on their own identify someone. The ICO says it "is sometimes known as the mosaic or jigsaw effect", and that simply removing direct identifiers does not make anonymisation effective19. Linked business records, such as a ticket joined to its email thread, carry the highest risk.

More: anonymisation standard.

M

Members' voluntary liquidation

A solvent winding up, available when the directors make a statutory declaration that, after full inquiry, the company will pay its debts in full with interest within a stated period of no more than 12 months6,7. It is used to close a solvent company with assets to distribute to shareholders, and the liquidator can realise those assets first, including by licensing the company's records.

More: closing your company.

Motivated intruder test

The ICO's yardstick for apparently anonymous data: could someone motivated to identify people succeed, using all practical steps and means reasonably likely to be used19? The ICO says you must use it to assess identifiability risk19. For company records, assume the intruder includes well-informed insiders, such as former colleagues, customers and suppliers, who know far more than a stranger does.

More: anonymisation standard.

O

Office-holder

The insolvency practitioner holding office in a company's insolvency: the administrator, administrative receiver, liquidator or provisional liquidator8. The office-holder acts for the company and has statutory powers to sell or dispose of its property, so decides whether its records are licensed and on what terms40,41. Payments to them from an estate are governed in England and Wales by SIP 929.

More: office-holder guide.

P

Personal data

Any information relating to an identified or identifiable person, who can be identified directly or indirectly, for example by a name, an identification number, location data, an online identifier or factors specific to their physical, economic, cultural or social identity12. Business records are full of it: email headers, signatures, job titles, phone numbers and free text about colleagues and customers.

More: anonymisation standard.

Pseudonymisation

Processing personal data so that it can no longer be attributed to a specific person without additional information kept separately and protected12. Replacing names with consistent codes is the common example. The ICO is clear that pseudonymous data is still personal data, because pseudonymisation reduces the links between people and their data without removing them18. It is a useful safeguard, not an exit from UK GDPR.

More: anonymisation standard.

R

Recognised legitimate interests

A lawful basis added to UK GDPR by the Data (Use and Access) Act 2025, available only for purposes listed in a new Annex 115. The list covers disclosures for public tasks, national security, public security and defence, emergencies, crime, and safeguarding vulnerable people16. Commercial licensing to AI developers is not on it, so sellers still rely on ordinary legitimate interests or another basis.

More: UK law guide.

Referral fee

A payment a buyer makes to whoever introduced a company that goes on to license its data. micro1, for example, publishes USD 50,000 on enterprise deals up to USD 700,000 and 7.5% above that, paid once the data is transferred and approved2. A seller's agent paid by the buyer must say so: the Supreme Court has said full disclosure and fully informed consent protect a fiduciary38. Sleeping Asset is paid this way, by buyers only, and discloses each fee in writing before the seller signs.

More: how we are paid.

Re-identification

Taking steps that make de-identified personal data identifiable again. In the UK it is a criminal offence to do so knowingly or recklessly without the consent of the controller responsible for de-identifying it, subject to defences such as preventing crime or the public interest21. A licence should also forbid re-identification and linking, so the buyer and anyone it passes data to are bound by contract too.

More: anonymisation standard.

Reinforcement learning environment

A practice setting in which an AI model is given a task and a set of actions, attempts the task, and is scored by a grader so that successful strategies are reinforced. The environment is the available actions plus the surrounding context, such as file systems or a simulated application34. Epoch AI reports growth in enterprise workflow tasks such as navigating Salesforce34. Also called an RL environment.

More: what AI companies buy.

Restricted transfer

The ICO's term for sending personal data to a separate organisation outside the UK, where UK GDPR applies to the data and you initiate the transfer20. Every restricted transfer must rely on UK adequacy regulations, an appropriate safeguard such as the international data transfer agreement, or a narrow exception20. Removing personal data inside the UK, before anything goes to a buyer abroad, avoids the question.

More: UK law guide.

S

SIP 9

Statement of Insolvency Practice 9, the professional standard on payments to office-holders and their associates from an estate in England and Wales, effective from 1 April 202129. It asks office-holders to disclose all payments arising from an appointment to themselves or their associates, treating as an associate anyone a reasonable and informed third party might see as one29. Creditors will expect agents' fees to be disclosed too.

More: office-holder guide.

Special category data

Personal data revealing racial or ethnic origin, political opinions, religious or philosophical beliefs or trade union membership, plus genetic data, biometric data used to identify someone, health data and data about sex life or sexual orientation14. Processing it is prohibited unless a specific condition applies14. Sickness records, occupational health files and biometric clocking data should be left out of any licensing project.

More: anonymisation standard.

Striking off

Removing a company from the register at Companies House. On the directors' application, the registrar must publish a notice in the Gazette and wait at least two months for objections, and the company is dissolved when notice of the striking off is published10. Anything the company still owns at that point becomes bona vacantia11, so records worth licensing should be dealt with first.

More: closing your company.

Synthetic data

Data generated by models or simulations rather than recorded from real activity. Buyers increasingly use real records to seed it: micro1 says some datasets "may undergo synthetic rewrites that preserve workflow structure"1. Research in Nature found that "indiscriminate use of model-generated content in training causes irreversible defects", which the authors call model collapse36. That supports demand for real records, without guaranteeing prices.

More: what AI companies buy.

T

Text and data mining

Using computers to analyse large volumes of text or data. UK copyright law allows copies for computational analysis only by someone with lawful access and "for the sole purpose of research for a non-commercial purpose"22. In March 2026 the government said a broad exception with opt-out "is no longer the government's preferred way forward"25. Commercial AI training on private records therefore needs a licence.

More: UK law guide.

U

UK data bridge

The UK's adequacy arrangement with the United States, in force since 12 October 2023, which lets personal data go to US organisations certified to the UK Extension to the EU-US Data Privacy Framework without further safeguards26,27. The recipient must appear on the Data Privacy Framework List, and HR data needs a certification that covers it27. Other US recipients need safeguards or an exception.

More: UK law guide.

FAQ

Questions about AI data licensing terms

What is AI data licensing?

AI data licensing lets an AI developer, or a data company that supplies developers, use a defined copy of an organisation's records to train or test models, on written terms and usually for a fee. The organisation keeps ownership. Personal data should be removed, preferably before records leave the organisation, and its own solicitor should approve the licence.

What is the difference between anonymisation and pseudonymisation?

Anonymisation reduces the risk of identifying anyone to a sufficiently remote level, so the result falls outside data protection law. Pseudonymisation, such as swapping names for codes, only separates identifiers from the data; the means of linking back still exist. Pseudonymised data remains personal data under UK GDPR; effectively anonymised data does not, though creating it is still processing18.

What is a data manifest?

A data manifest is a metadata-only summary a company completes for a prospective buyer, describing its systems, the records they hold, the years covered and rough volumes. It lets the buyer judge fit before any records are shared. micro1 and Cognor Labs ask for one at the start, and Cognor says its form takes 10 to 15 minutes2,31.

What is an RL environment?

An RL environment, short for reinforcement learning environment, is a simulated setting in which an AI model practises a task, such as resolving a ticket in a rebuilt business system, and is scored so that good strategies are reinforced34. Developers increasingly build these around real business workflows, which is why they want companies' email, tickets and system records.

Sources

  1. 1micro1, Monetize your company's operational data (data partnerships page and FAQ)micro1.ai · Read 12 Nov 2026
  2. 2micro1, Refer a company to micro1, earn USD 50,000+ (process, partnership types and FAQ)micro1.ai · Read 12 Nov 2026
  3. 3Insolvency Act 1986, Schedule B1, paragraph 3 (purpose of administration)legislation.gov.uk · Read 12 Nov 2026
  4. 4Insolvency Act 1986, section 122 (circumstances in which company may be wound up by the court)legislation.gov.uk · Read 12 Nov 2026
  5. 5Insolvency Act 1986, section 136 (functions of official receiver in relation to office of liquidator)legislation.gov.uk · Read 12 Nov 2026
  6. 6Insolvency Act 1986, section 89 (statutory declaration of solvency)legislation.gov.uk · Read 12 Nov 2026
  7. 7Insolvency Act 1986, section 90 (distinction between members' and creditors' voluntary winding up)legislation.gov.uk · Read 12 Nov 2026
  8. 8Insolvency Act 1986, section 234 (getting in the company's property; meaning of office-holder)legislation.gov.uk · Read 12 Nov 2026
  9. 9Insolvency Act 1986, section 388 (meaning of act as insolvency practitioner)legislation.gov.uk · Read 12 Nov 2026
  10. 10Companies Act 2006, section 1003 (striking off on application by company)legislation.gov.uk · Read 12 Nov 2026
  11. 11Companies Act 2006, section 1012 (property of dissolved company to be bona vacantia)legislation.gov.uk · Read 12 Nov 2026
  12. 12UK GDPR, Article 4 (definitions)legislation.gov.uk · Read 12 Nov 2026
  13. 13UK GDPR, Article 6 (lawfulness of processing)legislation.gov.uk · Read 12 Nov 2026
  14. 14UK GDPR, Article 9 (processing of special categories of personal data)legislation.gov.uk · Read 12 Nov 2026
  15. 15Data (Use and Access) Act 2025, section 70 (lawfulness of processing)legislation.gov.uk · Read 12 Nov 2026
  16. 16Data (Use and Access) Act 2025, Schedule 4 (Annex 1: recognised legitimate interests), in force 5 Feb 2026legislation.gov.uk · Read 12 Nov 2026
  17. 17GOV.UK, Data (Use and Access) Act 2025: plans for commencement, last updated 12 Oct 2026gov.uk · Read 12 Nov 2026
  18. 18ICO, Introduction to anonymisationico.org.uk · Read 12 Nov 2026
  19. 19ICO, How do we ensure anonymisation is effective?ico.org.uk · Read 12 Nov 2026
  20. 20ICO, A brief guide to international transfers, 15 Jan 2026ico.org.uk · Read 12 Nov 2026
  21. 21Data Protection Act 2018, section 171 (re-identification of de-identified personal data)legislation.gov.uk · Read 12 Nov 2026
  22. 22Copyright, Designs and Patents Act 1988, section 29A (copies for text and data analysis for non-commercial research)legislation.gov.uk · Read 12 Nov 2026
  23. 23Copyright, Designs and Patents Act 1988, section 90 (assignment and licences)legislation.gov.uk · Read 12 Nov 2026
  24. 24Copyright, Designs and Patents Act 1988, section 92 (exclusive licences)legislation.gov.uk · Read 12 Nov 2026
  25. 25GOV.UK, Report on Copyright and Artificial Intelligence, 18 Mar 2026gov.uk · Read 12 Nov 2026
  26. 26The Data Protection (Adequacy) (United States of America) Regulations 2023, SI 2023/1028legislation.gov.uk · Read 12 Nov 2026
  27. 27GOV.UK, UK-US data bridge: factsheet for UK organisationsgov.uk · Read 12 Nov 2026
  28. 28Civil Procedure Rules, Part 31 (disclosure and inspection of documents), rules 31.2 and 31.22justice.gov.uk · Read 12 Nov 2026
  29. 29ICAEW, Statement of Insolvency Practice 9 (England and Wales), Payments to insolvency office holders and their associates from an estate, effective 1 Apr 2021 (PDF)icaew.com · Read 12 Nov 2026
  30. 30Scale AI, Data Partnershipsscale.com · Read 12 Nov 2026
  31. 31Cognor Labs, homepage and FAQcognorlabs.com · Read 12 Nov 2026
  32. 32Labelbox Prism, FAQsprism.labelbox.com · Read 12 Nov 2026
  33. 33Endless Data, Terms and Conditions for its data customers, last updated 10 Jan 2025endless-data.com · Read 12 Nov 2026
  34. 34Epoch AI, Jean-Stanislas Denain and Chris Barber, An FAQ on reinforcement learning environments, 12 Jan 2026epoch.ai · Read 12 Nov 2026
  35. 35The Insolvency Service, Business insolvency demography 2015 to 2025: data tables, Table B3a (xlsx)assets.publishing.service.gov.uk · Read 12 Nov 2026
  36. 36Shumailov and others, AI models collapse when trained on recursively generated data, Nature, 24 Jul 2024nature.com · Read 12 Nov 2026
  37. 37Information Commissioner's Office, ICO takes enforcement action against Experian after data broking investigation, 28 Oct 2020 (press release as republished by Wired-Gov)wired-gov.net · Read 12 Nov 2026
  38. 38UK Supreme Court, Hopcraft and another v Close Brothers Ltd and related appeals [2025] UKSC 33, press summary, 1 Aug 2025supremecourt.uk · Read 12 Nov 2026
  39. 39ICO, Data sharing: a code of practice, Due diligence when sharing data following mergers and acquisitionsico.org.uk · Read 12 Nov 2026
  40. 40Insolvency Act 1986, Schedule 1 (powers of administrator or administrative receiver)legislation.gov.uk · Read 12 Nov 2026
  41. 41Insolvency Act 1986, Schedule 4 (powers of liquidator in a winding up)legislation.gov.uk · Read 12 Nov 2026
  42. 42Insolvency Act 1986, section 389 (acting without qualification an offence)legislation.gov.uk · Read 12 Nov 2026
  43. 43The Insolvency Service, Company Insolvency Statistics September 2026: data tables (Table 1a)assets.publishing.service.gov.uk · Read 12 Nov 2026

Every source was read on the date shown. Buyers change their terms often; if you spot something out of date, email support@sleepingasset.com and we will correct it.