A rights checklist before offering a dataset

Ask five different questions

Can you access the records? Who owns the relevant intellectual property? What rights did your organization receive? What promises limit disclosure? What privacy or sector-specific rules apply to the intended use? These questions overlap, but a yes to the first does not answer the others.

Write an inventory of the proposed dataset’s components and the evidence for each answer. Treat source code, text, photographs, recordings, metadata, and labels separately where their histories differ. A buyer’s willingness to purchase is not a substitute for this review.

Build a chain of authority

Collect the agreements and business records showing how rights reached the proposed seller: employment terms, contractor assignments, acquisition documents, contributor terms, supplier licenses, or relevant customer contracts. In U.S. copyright law, possession of a copy does not itself convey copyright ownership.

Make uncertain components visible. A missing contractor agreement, an unrecorded acquisition schedule, or a field supplied by a third party should become an open review item. Avoid replacing missing evidence with a blanket statement that the company owns everything in its systems.

Match the evidence to the proposed use

A license allowing internal analytics may not answer a request for commercial model training or sublicensing. Ask the buyer to describe the actual use, users, affiliates, derivative artifacts, and intended retention. Give the reviewer that description alongside the source agreements.

Review privacy and confidentiality independently of copyright. Even an original company-written document can contain another party’s confidential information or personal details. Removing the third party’s name does not necessarily remove the obligation or the ability to identify them from context.

• Record the exact entity that will grant the license.
• List excluded sources and unresolved permission gaps.
• Keep supporting evidence tied to dataset versions.

Turn review into an operational gate

Assign an owner to clear each issue before material enters a sample or delivery. Maintain three states: approved for the stated use, excluded, and pending. Do not allow pending material into a buyer evaluation simply because the full commercial agreement has not yet been signed.

When the use, recipient, geography, or dataset changes, reopen the relevant review. A clearance decision should describe its scope and assumptions rather than look like an unlimited certificate. This is a decision-support checklist; qualified counsel should resolve legal uncertainty for the actual transaction and jurisdictions.