Overview

AB 2013 asks developers to describe more than data origins: whether material may be protected or public domain, whether datasets were purchased or licensed, whether personal or aggregate consumer information may be present, how data was processed, and whether synthetic-data generation was used. These fields describe different branches of the path from dataset to model: legal status is not acquisition route, privacy controls do not prove absence, and processing or synthetic-data purpose does not establish an outcome. OpenAI and xAI illustrate how qualifiers such as “may be protected,” “may include,” “incidentally,” and “necessary rights” keep provider representations distinct from independent legal, privacy, quality, or safety conclusions.

Learning outcomes

Public-domain status, purchase, and licensing describe different properties and should not be collapsed into a generic rights claim.
Privacy disclosures identify possible presence and described controls; they do not prove personal information is absent or risk eliminated.
Processing and synthetic-data statements describe development purposes or use, not independently measured gains in quality, safety, or performance.

Keywords

AB 2013rightsprivacyprocessingsynthetic datalicensing

What the Data Statement Says About Rights, Privacy, and Processing

California AB 2013 treats a training-data disclosure as more than a story about where data came from. It asks a developer to describe the legal-status and acquisition conditions attached to that data, whether personal or aggregate consumer information may be present, how the developer cleaned or processed the material, and whether synthetic-data generation entered development. These categories let a reader examine the path from dataset to model without confusing a provider’s description of that path with an independent legal or privacy guarantee.

The categories are connected, but they answer different questions. A dataset can include material that may be protected by copyright and also include material in the public domain. It can be purchased or licensed without being entirely public domain. It can be processed to improve a model while still containing information the developer says may be personal. It can include synthetic data without that fact proving anything about the quality or safety of the resulting system.

Rights status is not the same as acquisition path

Section 3111 asks whether the datasets include data protected by copyright, trademark, or patent, or whether the datasets are entirely in the public domain. It separately asks whether the datasets were purchased or licensed. The separation is important because legal status and acquisition pathway describe different properties.

“Public domain” is a statement about the status of material. “Purchased” or “licensed” describes how a developer obtained or was permitted to use some material. The two ideas can overlap, but neither one automatically answers the other. A developer can report that a dataset contains public-domain material without telling the reader that the dataset was licensed. A licensed dataset can contain material that is not entirely in the public domain. The statutory structure prevents these categories from being collapsed into the single reassuring phrase “we have rights.”

OpenAI’s public summary illustrates the first distinction through a qualification. OpenAI says its diverse development data includes data that may be protected by copyright and data in the public domain. The phrase “may be protected” should remain attached to the claim. It reports how OpenAI characterizes the possibility; it does not identify the legal status of every item or resolve a dispute over a particular work.

xAI’s disclosure uses a different formulation. It says its datasets include data in the public domain and datasets that xAI has the necessary rights to use for training purposes, including because certain datasets were purchased or licensed. That statement gives the reader both a status category and examples of an acquisition route. It remains xAI’s representation of its rights position, not an independent ruling on copyright, trademark, or patent law.

Published by DeepMox

FAQ

Why does AB 2013 separate legal status from purchase or licensing?

Because legal status and acquisition pathway are different properties: public-domain status describes the material, while purchase or licensing describes how it was obtained or permitted for use. They can overlap, but neither answers the other.

Does an opt-out or response-removal request prove that personal information was deleted from training data?

No. OpenAI’s reported opt-out concerns users’ content being used for training, while its response-removal request concerns certain personal information in ChatGPT responses; the source does not say either deletes every corresponding item from every training dataset.

What does synthetic-data use establish about a model?

It establishes reported use and the provider’s stated purpose, not a quality guarantee. The cited provisions do not show that synthetic data is accurate, representative, unbiased, safer, or effective.

Previous10 minOpenAI and xAI: AB 2013 Disclosures

Next9 minAB 2013: Training-Data Timeline

AB 2013: Rights, Privacy, and Processing

What the Data Statement Says About Rights, Privacy, and Processing

Rights status is not the same as acquisition path

FAQ