Learn
Glossary of data licensing terms
Plain definitions of the terms that come up when a company licenses its records for AI training, including bankruptcy and privacy terms. General information, not legal advice.
- ABC (assignment for the benefit of creditors)
- A state-law alternative to bankruptcy in which a company assigns its assets to an assignee, who sells them and distributes the proceeds to creditors.
- 363 sale
- A sale of a debtor's assets outside the ordinary course of business in a US bankruptcy, approved by the court under 11 U.S.C. § 363.
- Consumer privacy ombudsman
- A person a US bankruptcy court may appoint before approving a sale of personally identifiable information when the debtor's privacy policy restricts its transfer (11 U.S.C. § 332).
- De-identification
- Removing or replacing personal identifiers (names, emails, phone numbers, addresses, IDs) so records can't reasonably be linked back to a person. FileFunding's illustrations replace them with role labels such as [Customer].
- PII (personally identifiable information)
- Information that identifies a person directly or in combination, such as a name, email address, phone number or account ID.
- Data license
- A contract giving a buyer the right to use a copy of data for agreed purposes, such as AI training, while the owner keeps the records. Most business data deals are licenses rather than outright sales.
- Exclusive license
- A license under which the owner agrees not to license the same data to anyone else, usually for a higher price. A non-exclusive license allows other licenses.
- Provenance
- The documented history of a dataset: who owned it, how it was collected and exported, what was excluded, and who had authority to license it.
- Provenance pack
- The documents that travel with a data lot: authorization, source systems and date ranges, inclusions and exclusions, applicable privacy notices, de-identification method and use restrictions.
- Training data
- Text, code, audio, video or other records used to train or fine-tune an AI model.
- Workflow traces
- Recordings or logs of multi-step tasks as people actually performed them, used to train computer-use agents.
- RL environment
- A simulated task setting, often built from real workflows, where an AI agent learns by trial and feedback (reinforcement learning).
- Data lot
- A defined, scoped set of records from one company offered for license, with its own provenance pack.
- Indicative estimate
- A rough range from public data points and a few answers. Not an offer, appraisal or valuation.