AI training data
AI training data is the collection of examples a machine learning model learns from. For a language model that means text, code and conversations; for a robot it means video and sensor readings paired with actions. The model adjusts itself until its outputs match the patterns in those examples.
Anonymization
Anonymization is processing data so the people in it cannot be identified by any means reasonably likely to be used, by anyone. Truly anonymous data falls outside the GDPR, but the bar is high: if re-identification is reasonably possible, the data is only pseudonymized.
Assignment for the benefit of creditors (ABC)
An assignment for the benefit of creditors is a state-law alternative to bankruptcy in which a failing company transfers its assets to an assignee, who sells them and pays creditors from the proceeds. Company data can be one of the assets the assignee sells or licenses.
Company data inventory
A company data inventory is a list of the record types a business holds, with the system each lives in, the years covered, approximate volume, owner and any restrictions. It is the first document a data buyer needs and can be built without sharing a single file.
Computer-use agent
A computer-use agent is an AI model that operates software the way a person does: it reads the screen, moves the cursor, clicks and types to complete a task. Training and testing these agents needs recordings and logs of real people doing real work in real applications.
Data licensing
Data licensing is granting another party permission to use data you control, for a defined purpose, period and price, while you keep ownership. In AI, a company typically licenses records so a developer can train or evaluate models on them. The license, not the file transfer, sets what the buyer may do.
Data licensing agreement
A data licensing agreement is the contract that sets what a licensee may do with licensed data: the dataset covered, permitted and prohibited uses, exclusivity, territory and term, privacy and security duties, retention and deletion, warranties about rights, liability limits and payment.
Data provenance
Data provenance is the documented history of a dataset: where each record came from, who created it, under what terms it was collected, and every transformation since. In AI licensing, provenance is how a buyer confirms the seller has the right to license the data.
Dataset card
A dataset card is a short document that describes a dataset: what it contains, how and when it was collected, what was removed, known gaps and biases, intended uses and the license. It lets a buyer judge fit without seeing the raw records.
De-identification
De-identification is removing or transforming information in a dataset so the people in it can no longer be readily identified. Methods range from deleting direct identifiers such as names and emails to generalizing dates and locations. Whether the result counts as de-identified depends on the law that applies.
Egocentric video
Egocentric video is footage recorded from the point of view of the person doing a task, usually with a head-mounted or chest-mounted camera. Robotics and physical AI developers use it to teach models how hands, tools and objects move in real work.
Evaluation dataset
An evaluation dataset is a set of examples kept out of training and used only to measure how well a model performs. Because it must stay unseen, buyers value fresh, private evaluation data that no model has trained on.
Exclusive data license
An exclusive data license gives one licensee the sole right to use a dataset for the licensed purpose, so the owner cannot license the same data to anyone else for that purpose during the term. Exclusivity can be full, or limited by use, field, territory or time.
Fine-tuning
Fine-tuning is further training of an already-trained model on a smaller, specialized dataset so it performs better on a particular task or domain. Company records are often licensed for fine-tuning rather than for training a model from scratch.
Non-exclusive data license
A non-exclusive data license lets a licensee use a dataset while the owner stays free to license the same data to other parties. Each deal usually pays less than an exclusive one, but the same records can earn from several buyers.
Personally identifiable information (PII)
Personally identifiable information is any information that can identify a specific person, alone or combined with other data: names, emails, phone numbers, account numbers, faces, voices and many indirect details. The GDPR and US state laws define the related terms "personal data" and "personal information" more broadly still.
Preference data
Preference data records which of two or more model responses a person judged better, and often why. It is used to train models to follow instructions and match human judgment, most famously through reinforcement learning from human feedback (RLHF).
Pseudonymization
Pseudonymization replaces identifiers with codes or tokens so data can no longer be linked to a person without extra information that is kept separately. Under the GDPR, pseudonymized data is still personal data, because the link can be restored.
Retrieval-augmented generation (RAG)
Retrieval-augmented generation is a technique where an AI system looks up relevant documents at answer time and uses them to write its response, instead of relying only on what it learned in training. Data licensed for retrieval is read, not learned.
RL environment
An RL environment is a simulated setting where an AI agent takes actions, observes results and receives a reward for reaching a goal, so it can learn by trial and error. For AI agents, environments imitate real software and workplaces, such as a ticketing system with realistic tickets.
Section 363 sale
A Section 363 sale is a sale of a bankrupt company's assets, often by auction, approved by the bankruptcy court under 11 U.S.C. § 363. If the assets include personally identifiable information that the company's privacy policy promised not to transfer, the court can approve the sale only on conditions, typically after a consumer privacy ombudsman reviews it.
Section 7216
Section 7216 of the US Internal Revenue Code makes it a crime for a tax return preparer to knowingly or recklessly disclose or use a client's tax return information without permission. Treasury regulations allow some uses with the client's specific written consent.
Teleoperation data
Teleoperation data is recorded while a person remotely controls a robot to perform a task. It captures the robot's camera views together with the exact movements the operator commanded, which makes it a direct example for the robot to imitate.
Reviewed November 7, 2026 by Origin Data Partners.