California regulates AI training data at four points: the CCPA treats licensing personal information for money as a sale, CPPA regulations require a risk assessment before that sale, AB 2013 makes generative AI developers publish whether their training data was purchased or licensed, and CIPA requires every party's consent to record a confidential call.
Status is given as of November 9, 2026. A seller holding Californians' data meets the sale, assessment and recording rules, while the buyer meets AB 2013, which is in force while a constitutional challenge to it is on appeal. Together they form the California layer of AI training data licensing, and the sections below follow the order of a deal.
Four California Rules on a Training Data Deal at a Glance
Four California instruments decide what a seller may license and what the buyer later discloses, and the table gives each one's citation, start date and status as of 2026-11-09.
| Instrument | Citation | In force | Who it binds | What it requires in a data deal | Status | Source |
|---|---|---|---|---|---|---|
| California Consumer Privacy Act (as amended by the CPRA) | Cal. Civ. Code 1798.100 to 1798.199.100 | 2020-01-01; CPRA amendments 2023-01-01 | businesses above the thresholds | licensing personal information for money is a sale; opt-out (1798.120); deidentified data outside (1798.140(m)) | in force; statute version effective 2026-01-01 | cppa.ca.gov |
| CPPA regulations on risk assessments, cybersecurity audits and ADMT | Cal. Code Regs. tit. 11, 7150 to 7157, 7200 | 2026-01-01 | businesses subject to the CCPA | risk assessment before selling personal information or training ADMT on it; section 7153 facts to the recipient | in force; OAL approval 2025-09-22 | cppa.ca.gov |
| AB 2013, Generative Artificial Intelligence: Training Data Transparency | AB 2013 | 2026-01-01 | developers of generative AI | public high-level summary of training datasets, including whether data was purchased or licensed and whether it holds personal information | in force; challenge on appeal, 9th Cir. No. 26-1591 | Ninth Circuit amicus filings |
| California Invasion of Privacy Act | Cal. Penal Code 630 to 638.55 (631, 632) | 1967 | anyone recording or intercepting communications | all-party consent to record confidential communications (632); no wiretapping (631) | in force | federal court orders quoting the sections |
The sections below take the checkpoints in deal order: sale, deidentification, assessment, disclosure and recordings.
When Licensing Personal Information for AI Training Is a Sale Under the CCPA
Licensing personal information to an AI developer for money is a sale under the California Consumer Privacy Act (Cal. Civ. Code 1798.100 to 1798.199.100), so the consumers concerned keep the right to opt out of it (1798.120). The CCPA has been in force since 2020-01-01, the CPRA amendments apply from 2023-01-01, and the California Privacy Protection Agency (CPPA) posted the statute version effective 2026-01-01. Rules on resale notices and on passing deidentification duties to recipients also apply to the seller.
The statute treats an AI system capable of outputting personal information as a format of personal information, so a trained model that can output it can itself hold it.
Which Businesses the CCPA Covers: Revenue and Volume Thresholds
The CCPA binds a for-profit business that crosses one of its thresholds, and two of them matter for a data seller:
- annual gross revenue above USD 25,000,000 as adjusted, which the CPPA CPI-adjusted to USD 26,625,000 from 2025-01-01 (1798.140(d)(1)(A));
- buying, selling or sharing the personal information of 100,000 or more consumers or households.
A third threshold exists in the statute and is not described here.
Staff and Business Contact Records Since January 2023
Employee records and business contact data count as personal information in a sale, because the CCPA exemptions for both stopped operating on January 1, 2023 (1798.145(m)(4) and (n)(3)). Slack, email or HR archives about California staff therefore fall under the sale and opt-out rules like customer records. The notice a company owes its staff, and the comparison with other countries, sit on the pages for employee notice and consent when licensing internal communications and the employee rules.
Deidentified Data Under Civil Code 1798.140(m): Three Conditions
Deidentified data falls outside the CCPA's personal information when the business meets all three conditions of Civil Code 1798.140(m):
- it takes reasonable measures so the data cannot be associated with a consumer;
- it publicly commits not to reidentify the data;
- it places contractual obligations on every recipient to do the same.
The third condition makes deidentification a contract term between seller and buyer, not only a technical step. The methods for PII removal and de-identification before licensing data are covered on the seller page.
A Court-Supervised Data Sale That Adopted the CCPA Test
The Spirit Airlines chapter 11 estate (Bankr. S.D.N.Y. 25-11897 (SHL)) offers its internal business data only after de-identification to the CCPA standard, with 45 CFR 164.514 applied to health data (Form of Bill of Sale, Dkt. 1463, 2026-08-14). An independent Deidentification Agent, Tonic.ai, performs the work, and the consumer privacy ombudsman's report (Dkt. 1684, 2026-10-05) records that the debtors adopted the 1798.140(m) standard by letter of 2026-08-05.
The sale is not approved as of November 9, 2026, and the sale hearing is set for 2026-11-17 (Dkt. 1779).
A transfer of personal information as an asset in a bankruptcy or similar transaction in which the transferee assumes control of all or part of the business is not a CCPA sale (1798.140(ad)(2)(C), (ah)(2)(C)), which is the route described under US bankruptcy data sales under Section 363(b)(1).
Risk Assessments Before Selling Personal Information: The 2026 CPPA Regulations
Since January 1, 2026, CPPA regulations require a business to complete a risk assessment before it sells personal information, and before it processes personal information it intends to use, or let others use, to train automated decisionmaking technology (ADMT) for significant decisions or facial recognition, emotion recognition or identity verification technology (Cal. Code Regs. tit. 11, 7150 to 7157, 7200). The CPPA board adopted the text on 2025-07-24 and the Office of Administrative Law approved it on 2025-09-22.
The table lists the dates and periods a seller tracks, as of 2026-11-09.
| Date or period | Rule | Source |
|---|---|---|
| 2026-01-01 | regulations effective; risk assessment before selling personal information | cppa.ca.gov/regulations/ccpa_updates.html |
| 2027-01-01 | ADMT compliance date for significant decisions | Cal. Code Regs. tit. 11, approved text (cppa.ca.gov) |
| 2027-12-31 | deadline to assess processing that began before 2026-01-01 | Cal. Code Regs. tit. 11, approved text (cppa.ca.gov) |
| 2028-04-01 | submission deadline for 2026 to 2027 risk assessments | Cal. Code Regs. tit. 11, approved text (cppa.ca.gov) |
| within 45 calendar days | update after a material change in the processing | Cal. Code Regs. tit. 11, approved text (cppa.ca.gov) |
| 5 years | retention after completion or for the duration of processing, whichever is later | Cal. Code Regs. tit. 11, approved text (cppa.ca.gov) |
Section 7153 makes a business that provides ADMT trained on personal information to another business give that business the facts it needs for its own assessment, which a license should allocate, as the page on AI data licensing agreements explains clause by clause. ADMT used for hiring decisions is a deployer topic and is not covered here.
AB 2013: What AI Developers Must Publish About Licensed Training Data
AB 2013, California's Generative Artificial Intelligence: Training Data Transparency law in force since January 1, 2026, requires developers of generative AI to post a high-level summary of their training datasets, including whether data was purchased or licensed and whether it contains personal information. A seller's data category therefore becomes publicly visible.
A brief filed in the Ninth Circuit lists twelve categories:
- sources or owners of the datasets;
- the purpose of the datasets;
- the number of data points;
- the types or labels of data;
- copyright, trademark or patent status;
- whether the data was purchased or licensed;
- whether it contains personal information;
- whether it contains aggregate consumer information;
- cleaning or processing of the data;
- the collection period;
- the dates the datasets were first used;
- whether synthetic data was used.
The EU training-content summary, by contrast, discloses licensed datasets only as Yes or No plus modality, as set out under EU AI Act Article 53 training data summary.
The Constitutional Challenge on Appeal (Ninth Circuit No. 26-1591)
AB 2013 stays in force while X.AI LLC challenges it: the federal district court in the Central District of California denied a preliminary injunction, and the appeal is pending in the Ninth Circuit as No. 26-1591, with no decision found as of November 9, 2026. The challenge to the Attorney General of California raises trade secret, takings and compelled speech arguments, and amicus briefs supporting California were filed on 2026-07-22 (DktEntry 28.1 and 36.2).
Recorded Calls and Meetings in a Data Sale: CIPA Sections 631 and 632
California's Invasion of Privacy Act (Cal. Penal Code 630 to 638.55, in force since 1967) requires the consent of all parties to record a confidential communication (632) and bars wiretapping (631), so a call or meeting archive recorded without that consent is a liability in any sale. Federal law allows one-party consent (18 U.S.C. 2511(2)(d)), but California does not.
AI vendors that use calls for their own training may be treated as third-party eavesdroppers, and under 631(a)(3) liability for later use needs a predicate violation of (a)(1) or (a)(2). Recording rules in other states are compared under call and meeting recording consent rules by jurisdiction.
Two federal orders of 2026 apply these sections to training: the Google contact center order of 2026-10-22 found no 631(a)(3) claim from training use of consented recordings without a predicate violation, and the Otter order of 2026-08-13 let a CIPA 631 claim proceed. Both appear in the table below.
California Court Records on Training Data, as of November 2026
Four California-law matters touch data used for AI training, and none of them has a final judgment on the merits as of November 9, 2026.
| Date | Court and docket | Matter | Outcome as stated in the record | Source | Page | Status |
|---|---|---|---|---|---|---|
| 2026-10-22 | N.D. Cal. 3:23-cv-05437-RFL, Dkt. 140, 141 | CIPA 631(a), 632(a), 637.5 claims over a contact center AI product | Verizon, Uber and Discover callers compelled to arbitration (Dkt. 140); Home Depot callers' 631(a) claims dismissed with leave to amend by 2026-11-12, accountholders consented via the Home Depot privacy statement, training use no 631(a)(3) claim without a predicate violation (Dkt. 141) | govinfo.gov | In re Google Cloud Contact Center AI Privacy Litigation | pending claims, amendment allowed |
| 2026-10-05 / 2026-08-14 | Bankr. S.D.N.Y. 25-11897 (SHL), Dkt. 1684, 1463 | data sale in a chapter 11 | Bill of Sale de-identifies to the CCPA standard, 45 CFR 164.514 for health data; ombudsman report records 1798.140(m) adopted; sale not approved, hearing 2026-11-17 (Dkt. 1779) | courtlistener.com (docket copies) | Spirit Airlines' Data Archive Sale in Bankruptcy | sale not approved |
| 2026-08-13 | N.D. Cal. 5:25-cv-06911-EKL, ECF 68 | speech recognition privacy suit | motion to dismiss granted in part; ECPA, CIPA 631, BIPA, UCL claims proceed; plaintiffs allege training on recorded conversations and voiceprints (allegation) | govinfo.gov | In re Otter.AI Privacy Litigation | claims proceed, no merits ruling |
| order date not stated | C.D. Cal.; 9th Cir. No. 26-1591 | challenge to AB 2013 | preliminary injunction denied; appeal pending, amicus briefs 2026-07-22; no decision found | Ninth Circuit amicus filings | X.AI v. Bonta | appeal pending |
Statuses come from the dockets and change with each filing.
Penalties and Enforcement for a California Data Seller
The CPPA can fine a business up to USD 2,500 per CCPA violation and USD 7,500 per intentional violation or violation involving personal information of consumers known to be under 16 (1798.155(a)), and the Attorney General can seek civil penalties of the same amounts (1798.199.90), both stated in the statute version effective 2026-01-01 and subject to CPI adjustment. The CPPA implements and enforces the CCPA. The CIPA claims in the cases above are brought by private plaintiffs.
Benefits, Risks, Downsides and Alternatives for a Seller Under California Law, Including Not Selling
California law lets a company license data for AI training, but personal information, staff records and recordings each add a condition, and licensing only deidentified data or not licensing at all remain valid choices.
Benefits
- Deidentified data meeting 1798.140(m) is outside personal information.
- An asset transfer in which the transferee assumes control of the business is not a sale.
- Businesses below the thresholds are outside the CCPA.
Risks
- Administrative fines per violation (1798.155(a)).
- CIPA claims over recordings made without all-party consent.
- The buyer's AB 2013 summary may show that the data was purchased or licensed.
Downsides
- Consumers keep an opt-out right over a sale (1798.120).
- A risk assessment comes before the sale and is kept for 5 years.
- Recipients must accept contract terms for deidentified data.
Alternatives
- License deidentified data only.
- Exclude recordings made without all-party consent and staff records.
- Wait for the Ninth Circuit outcome on AB 2013, which is not predicted here.
- Not selling or licensing at all.
The full range of alternatives to selling data to AI labs is covered on its own page.
Beyond California: Other States, Federal Rules and the Seller Process
California is one state layer on top of federal law. The United States AI training data law page compares it with eleven other states, while the seller pages turn these rules into steps.
- AI training data law outside the US: the jurisdiction overview.
- Staff data rules in other jurisdictions: employee data and AI training.
- Copyright in the licensed set: fair use and AI training in US law.
- What deidentified data achieves legally: anonymization vs pseudonymization for AI training data.
- The decision before any California sale: should you sell or license your data to AI companies.
- The steps of a sale: sell or license company data to AI companies.
- The CCPA beside the GDPR and other privacy laws: privacy law and AI training data compared.
- Biometric data in another state: Illinois BIPA and AI training data.
- The federal one-party rule: ECPA and the Wiretap Act in data sales.
- Health data de-identified under 45 CFR 164.514: HIPAA and AI training data.
- Call archives as a data type: selling call recordings for AI training.
How This Page Is Sourced and Dated
Every rule here cites the California code, the CPPA regulation text or a court docket, with the date it was checked. The methodology page explains how we source and date every rule. This page is general information about California law, not legal advice for any specific deal.
Frequently Asked Questions
Is it legal to sell data to AI companies in California?
Yes, with conditions. Licensing personal information for money is a CCPA sale with an opt-out right, a risk assessment comes first, and recordings need all-party consent.
Does the CCPA apply to AI training data?
Yes, whenever the set holds personal information of consumers of a covered business. That includes staff and business contact records since 2023-01-01.
Does AB 2013 make AI developers name the companies that licensed them data?
No. The summary is high-level and covers whether data was purchased or licensed, not who sold it.
Can a company license recorded customer calls for AI training in California?
Only with every party's consent to the recording. In the Google contact center order of 2026-10-22, training use of consented recordings was not a 631(a)(3) claim.