Review copy. Publication details are being completed. This copy is not available for contractual acceptance.
Version 2026-09-14.1 · Permanent link to this version · U.S. legal documents
Version history
- Version 2026-09-14.1 — Prepared 2026-09-14 · Draft — not published
- Version 2026-09-13.1 — Prepared 2026-09-13 · Draft — not published
1. What this page explains
Training-data transparency explains what information a developer used to build an AI system. It is different from the notice explaining how a hiring organization uses AI to assess an individual. For that information, read the candidate notices and the AI section of the Privacy Notice.
2. When California's training-data law applies
California Civil Code section 3111 requires covered developers to publish training-data information on their website before making a covered generative AI system, service or substantial modification publicly available to Californians. The duty began January 1, 2026 and covers releases from January 1, 2022 onward, whether use is paid or free. A developer includes a person or business that designs, codes, produces or substantially modifies the system or service for public use. Under section 3110, training includes the developer’s testing, validation and fine-tuning. Substantial modifications are updates that materially change functionality or performance.
Specific exemptions include a system or service whose sole purpose is security and integrity. That exemption does not cover a separate hiring-analysis purpose merely because the platform also provides security checks.
3. What Evidize and the hiring organization provide
Evidize must publish the required information when it is a covered developer. Another provider is responsible for its own covered development. Using that provider’s model does not, by itself, establish whether Evidize’s development, testing, validation or modifications are covered.
The hiring organization gives the notices required for its own use of AI. Ordinary use of a supplier’s model does not, by itself, make the organization responsible for publishing that supplier’s training-data disclosure. This page does not give permission to train on a person’s information.
4. Developer disclosure format
For a covered system or service, the developer publishes a high-level dataset summary containing the following information. The law permits ranges and estimates where stated below; it does not require publication of the underlying dataset.
| Topic | Information supplied by the developer |
|---|---|
| Origin | [Dataset sources or owners] |
| Purpose | [How the datasets support the intended use] |
| Size | [Number of data points or general range; estimated figures may be used for changing datasets] |
| Characteristics | [Label types, or general characteristics for unlabelled data] |
| Intellectual property | [Whether data are protected by copyright, trademark or patent, or are entirely in the public domain] |
| Acquisition | [Whether datasets were bought or licensed] |
| Personal information | [Whether present] |
| Aggregate consumer information | [Whether present] |
| Preparation | [Cleaning, processing or modifications and their purpose] |
| Collection dates | [Period and whether collection continues] |
| First use | [Dates first used in development] |
| Synthetic data | [Whether synthetic data generation was used or continues to be used in development] |
Bracketed fields show where the developer's facts belong. They are not statements that a particular dataset, training practice or exemption has been established.