Data Licence Review Tool: Risks Highlighted on the Dataset Licence
Upload a dataset or data-feed licence and read flags on permitted use, exclusivity, provenance, revocation and derived model ownership.
Checkory is a data licence agreement review tool for a dataset licence, a grant to use licensed data. On that file the permitted use for train, fine-tune or display, the exclusivity sentence, the provenance warranties, the revocation sentence and derived model ownership stay hard to see. The same dataset licence comes back with flags on permitted use, exclusivity, provenance warranties, revocation and derived model ownership, plus a short explanation and an England and Wales pointer.
A grant that allows display and still bars train or fine-tune, or a revocation sentence that reaches a model the team thought they owned, never gets a mark if the file is read as a processor paper or as a spreadsheet of usage rights. No fetched statute assigns the model to the licensee. The pass does not decide ownership. Before the team trains or displays the data, name which of those marks goes to a solicitor.
Data licence agreement review tool
The whole dataset or data-feed licence in one upload
Confirm the file is a dataset or data-feed licence, a grant to use licensed data, as a PDF or DOCX. A vendor processor agreement stays on the published processor page.
A grant to use licensed data
The Department for Science, Innovation and Technology report, published 18 March 2026, says a licence provides the rights to do acts restricted by copyright, including reproductions under section 17 of the Copyright, Designs and Patents Act 1988, and typically places conditions on the use of trained models. It may also require attribution, reporting or auditing. Section 16 lists those restricted acts. Infringement is doing or authorising them without the licence of the copyright owner. Do not decide that the word display is one of those acts, or that the word train is always section 17. The sentence on the file controls.
The same file comes back with flags
Open the upload at document analysis. A short explanation and an England and Wales pointer sit with the mark where a provision may apply.
What is a dataset licence versus a DPA
A dataset licence grants use of licensed data. A processor agreement is a different paper. The Information Commissioner data-sharing code, fetched 9 October 2026, covers personal data shared between controllers. Sharing with a processor is outside that code, and the code does not apply to information that is not personal data. Article 28 of the UK GDPR is the written contract: subject-matter, duration, nature and purpose, types of personal data, categories of data subjects, and processing only on documented instructions. One comparison is enough.
A dataset licence grants use of licensed data
Pros
- ✓Flags on the grant: permitted use, exclusivity, provenance, revocation and derived model ownership.
- ✓Display can stay open while train and fine-tune stay closed.
Cons
- ✗The pass does not decide section 92 or assign the model.
A processor agreement is a different paper
Pros
- ✓The honest upload for an Article 28 contract. Roles, sub-processors, transfers and deletion already have a published page.
Cons
- ✗Those marks do not show provenance, revocation, or who the file says owns the derived model.
A data-sharing score is not this comparison
Pros
- ✓A fair inventory when the job is usage rights, territory and term across a portfolio, or a score on a data sharing file.
Cons
- ✗A score, a workbook or a JSON payload is not flags on this dataset licence, and this page does not write a data sharing agreement.
Upload the dataset or data-feed licence
Drop the dataset or data-feed licence
Use the PDF or DOCX of the grant to use licensed data. A processor agreement and a software licence belong on their own pages.
Flags land on the grant
Read permitted use, exclusivity, the provenance warranty or its gap, revocation, and derived model ownership.
Analyse a contract at /document-analysis
The path is /document-analysis and the label is Analyse a contract. Take model ownership and revocation to a solicitor before the team trains or displays the data.
An Excel or JSON extract is not this upload
A workbook or a JSON catalogue payload is a different product.
High marks leave this pass
Send model ownership or revocation to a solicitor with the marked file before the team trains or displays the data. A revocation sentence that reaches a model the team thought they owned is a High mark. The pass does not decide ownership and does not replace a solicitor.
Use scope, exclusivity, provenance and exit on the dataset grant.
| Column on the grant | What the sentence does | What you do with the mark |
|---|---|---|
| Use scope | Names train, fine-tune, display, or a narrower purpose. Display can stay open while train and fine-tune stay closed. The sentence on the file controls. | Flag that sentence. Do not treat train as always section 17. |
| Exclusivity | Either shuts out everyone, including the person granting the licence, or it does not. The word exclusive on the PDF is not a finding that section 92 is met. | Flag the sentence and point at section 92. Do not decide it. |
| Provenance | No page fetched on 9 October 2026 requires a clause titled provenance warranty. Flag the warranty, or the gap. Do not invent one. | See whether the file warrants the source. A missing sentence stays a gap. |
| Exit | Shows what ends, and who the file says owns the model. No fetched statute assigns the model or requires unlearning. | Flag both sentences. If you cannot accept them, take the marked file to a solicitor. |
How the upload returns the same file
Drop the dataset or data-feed licence
Upload the grant you were sent. A training-rights questionnaire does not take the licence file. A repository scan of a public dataset URL is not this commercial grant.
Flags land on the grant
A neighbour review of a dataset or data-feed licence returns a Word report that quotes the clause. Their check counts stay on that page.
Analyse a contract at /document-analysis
Start at document analysis. The label is Analyse a contract.
An Excel or JSON extract is not this upload
One neighbour Contracko extracts usage rights, territory and term into a workbook, and the page asks whether AI training is allowed. The product is the workbook. Another returns the same fields as JSON from Contracko for a catalogue. The product is the payload.
See the marks on your dataset licence
Analyse a contractWhen model ownership or revocation needs a solicitor
Derived-model ownership on the marked file
The same report of 18 March 2026 says a statistical model sits at the heart of an AI system, and that pure statistical models and algorithms are not protected by copyright, because copyright does not extend to ideas, methods or mathematical concepts as such. Section 9(3) still says that for a computer-generated literary, dramatic, musical or artistic work the author is the person by whom the arrangements necessary for the creation of the work are undertaken. The report proposes removing that specific protection, while copyright should continue to protect works created with AI assistance. The proposal is not a repeal, and section 9(3) does not give the team the model.
Revocation and unlearning on exit
No fetched statute imposes a duty to unlearn a model. If the sentence requires deletion of the feed or of a model, flag that sentence. The report describes the licensing market as new and evolving and proposes not to intervene at this stage.
High marks go to a solicitor
Take model ownership or revocation to a solicitor with the marked file. The pass does not decide who owns the derived model.
What we mark on the dataset grant
Use scope
The row is the permitted-use sentence: train, fine-tune, display, or a narrower purpose. Section 16 lists the acts restricted by copyright. An improve or enhance clause in a vendor master services agreement is a separate how-to.
Exclusivity
Point at section 92 and do not decide that the cover word exclusive meets it. An exclusive licence means a licence in writing signed by or on behalf of the copyright owner, authorising the licensee to the exclusion of all other persons, including the person granting the licence. Regulation 23 of the Copyright and Rights in Databases Regulations 1997 applies sections 90 to 93 to database right.
Provenance
Flag the provenance warranty, or the gap where the file has none. Do not write a warranty the file does not contain. Section 3A says a database is a collection of independent works, data or other materials arranged in a systematic or methodical way and individually accessible. A literary work consisting of a database is original only if, by reason of the selection or arrangement of the contents, the database constitutes the author's own intellectual creation. Regulation 13 says database right can subsist where there has been a substantial investment in obtaining, verifying or presenting the contents, whether or not the database is a copyright work. Regulation 16 says extraction or re-utilisation of all or a substantial part without the consent of the owner infringes database right.
Exit
Read revocation and derived model ownership together. Ownership of outputs in a vendor contract is a separate how-to. This column does not make that how-to this page.
Dataset licence review tool
A Hugging Face or GitHub URL scan is not this page
A clear, caution or restricted label on a repository URL does not open the commercial licence in the inbox.
An SPDX or SBOM scanner is not this upload
A software bill of materials does not review this grant. Figures from a research model on public dataset licences stay off this page.
A training-rights questionnaire does not replace the licence file
Questions about source and a text-and-data-mining opt-out do not replace the sentence that allows display and still bars train or fine-tune.
This page marks the uploaded dataset licence
The file is the dataset or data-feed licence you were sent.
Data license agreement review tool
Same job as the EN-GB licence page
Data license agreement review tool is the US spelling of the same search. The slug and the title stay on the British spelling licence.
Contracko extracts a data sharing file
On the US-spelling results the data sharing review returns with the workbook and the JSON payload.
Justee reviews a processor agreement
A processor review against GDPR and CCPA appeared on the US-spelling search on 9 October 2026, not in the primary list. The fetched footer said Last updated October 9, 2026. Price and timing lines stay there.
Upload a data licence for review
A catalogue submission is not this review
Results for upload data licence for review are catalogue submissions. Mozilla Data Collective takes a dataset and a licence field. Discuss Data checks the licence before publication. Neither returns flags on the licence file.
The licence file is what gets uploaded
Upload the dataset or data-feed licence, PDF or DOCX, not the dataset and not a catalogue form.
Open the tool at /document-analysis
Open document analysis. The label is Analyse a contract.
Processor flags stay on the DPA page
Roles, sub-processors, transfers and deletion stay on the DPA page
Roles, sub-processors, transfers and deletion stay on the published processor tool.
A processor checklist stays on the blog
The live processor checklist stays on the blog. Transfer addenda and a breach clock stay in that cluster.
This comparison only separates a dataset licence from a DPA
The processor page is where a processor agreement goes.
A score or a processor pass is a different file
A data-sharing score, an Excel or JSON extract, or the published processor tool does not already cover this dataset licence. The score and the extract answer a portfolio or a different paper. The processor tool answers Article 28. Stay on the grant to use licensed data.
A training-rights how-to stays on the blog
One vendor MSA training clause stays on the live blog article
One vendor AI-training clause in a master services agreement is the live how-to, not this whole licence.
Train, fine-tune and display are one use-scope row
The row asks which of train, fine-tune and display the sentence names.
A TDM questionnaire is not this file
A screen of text-and-data-mining questions does not highlight the sentence on the grant.
Output ownership how-to stays on the blog
One vendor output-ownership clause stays on the live blog article
Ownership of AI outputs in a vendor contract is the live how-to. It is not this dataset licence.
Derived-model ownership is a mark on this dataset licence
Here that sentence sits beside revocation. Section 9(3) does not fill a silent line.
Software licence flags stay on the software page
Grant, metrics and audit of a software licence stay on that page
The published software licence page marks grant, metrics and audit.
A dataset licence is a different file
A dataset licence is not a software seat.
Data sharing stays off this page
A data sharing score stays off this page
A data sharing agreement is not written here. A score on that other file stays on that other product.
This page does not become a DSA versus DPA guide
The comparison only separates a dataset licence from a processor agreement. It does not grow into a data-sharing guide.
Chat, Word add-on and lifecycle software
A general chat is not this upload
A general chat does not put flags on the dataset licence.
A Word add-on is not this page
A Word add-on and a draft generator are not this page.
Contract lifecycle software is not this page
Contract lifecycle software does not replace this read of the grant.
This page returns flags on the uploaded dataset licence
After the pass you can name the permitted-use sentence, the exclusivity sentence, the provenance warranty or its gap, the revocation sentence, and the derived model ownership sentence. You can send model ownership or revocation to a solicitor with the marked file. You do not treat the pass as a solicitor opinion.
FAQ
Which file belongs in this dataset upload?
A dataset or data-feed licence, meaning a grant to use licensed data, as a PDF or DOCX. A vendor processor agreement, a software licence, a data sharing agreement and a catalogue submission of the dataset itself are other jobs. If the paper sets documented instructions for a processor, use the published processor page. If it prices seats and audits installs, use the software licence page.
Can display stay allowed while training stays barred?
Yes, if that is what the permitted-use sentence says. A grant that allows display and still bars train or fine-tune is a mark on the file. The sentence controls. Do not treat the word train as a finding that section 17 always applies, and do not treat the word display as a named restricted act. The report of 18 March 2026 says a licence typically places conditions on the use of trained models. Those conditions are whatever the file writes.
Does the word exclusive on the cover settle the grant?
No. Section 92 describes an exclusive licence in writing signed by or on behalf of the copyright owner, authorising the licensee to the exclusion of all other persons, including the person granting the licence. Regulation 23 applies sections 90 to 93 of the 1988 Act to database right. The word on the PDF is not a finding that the section is met. Point at the section and do not decide it.
What should I do when no provenance warranty is written?
Flag the gap. No page fetched on 9 October 2026 requires a clause titled provenance warranty. Do not invent one, and do not borrow a neighbour heading about source warranties. Regulation 16 is background only: extraction or re-utilisation of all or a substantial part without the consent of the owner infringes database right. The mark is the warranty or the missing sentence.
Does section 9(3) give our team the model?
No. The section, as fetched on 9 October 2026, names the author of a computer-generated literary, dramatic, musical or artistic work as the person by whom the arrangements necessary for the creation of the work are undertaken. The report of 18 March 2026 proposes removing that protection for wholly computer-generated works. The proposal is not a repeal, and the section does not assign the model to the licensee. No fetched statute imposes a duty to unlearn. If the revocation sentence requires deletion of the feed or of a model, flag that sentence and take it to a solicitor with the marked file.
Why not run a processor review or export the terms to a spreadsheet?
Those outputs answer a different job. A processor review marks roles, sub-processors, transfers and deletion. A spreadsheet or JSON extract inventories usage rights, territory and term. This pass flags the dataset grant you uploaded: permitted use, exclusivity, provenance warranties, revocation and derived model ownership. Model ownership and revocation that you cannot accept go to a solicitor with the marked file. The pass does not replace a solicitor.
Related guides
Related articles
