California AB 2013 AI Training Data Transparency

state_law — US-CA
Effective: January 1, 2026 56 platforms tracked 351 provisions indexed Enforced by: California Attorney General

Overview

California AB 2013 requires developers of generative AI systems to publicly disclose detailed information about the data used to train their models. Disclosures must include high-level summaries of training datasets, whether datasets include copyrighted material or personal information, the sources and methods of data collection, the number or volume of data points in training sets, and the time period covered by training data. The law aims to increase transparency around AI training practices, particularly regarding the use of copyrighted works and personal data in model development. Narrow exceptions exist for national security applications and certain trade secret protections.

Key Articles & Sections

Platforms We Track Subject to

Recent Changes Related to

ConductAtlas maps governance language to potentially relevant regulatory frameworks. Regulatory applicability and enforceability may vary by jurisdiction, enforcement context, and individual circumstances. This page is informational and does not constitute legal advice. Methodology

Provisions Governed by (351 across 56 platforms)

Acceptable Use Policy and Content Restrictions AI21 Labs
Medium
Unilateral Modification of Terms AI21 Labs
Medium
Subscription Auto-Renewal and Cancellation Deadline Anthropic
Medium
Intellectual Property Assignment of Outputs Anthropic
Medium
Election and Political Influence Restrictions Anthropic
Medium
Assignment of Output Ownership Anthropic
Medium
Automated Collection of Device and Location Data Anthropic
Medium
Model Training Opt-Out with Safety Review Carve-Out Anthropic
Medium
Content Ownership and Output Assignment Anthropic
Medium
Unilateral Terms Modification with 30-Day Notice Anthropic
Medium
Data Collection from Third-Party Training Sources Anthropic
Medium
Agentic AI and Model Context Protocol (MCP) Additional Guidelines Anthropic
Medium
Cookies and Similar Tracking Technologies Anthropic
Medium
Feedback Rights Assignment Anthropic
Medium
Misinformation and Deceptive Content Prohibition Anthropic
Medium
Agentic Use and Autonomous Action Guidelines Anthropic
Medium
Cookies and Tracking Technologies Anthropic
Medium
Broad Automatic Data Collection Anyscale
Medium
Prohibition on Representing Output as Human-Generated Cerebras
Medium
Prompt and Output Data Collection and Review Cohere
Medium
Intellectual Property in Content and Outputs Cohere
Medium
Scope of Terms and Services Covered Cohere
Medium
Prohibition on Influence Operations and Disinformation Cohere
Medium
Mandatory Use Case Approval Cohere
Medium
Unilateral Amendment of Terms Cohere
Medium
Harm Documentation Obligation Cohere
Medium
Prohibited Use: Election Interference and Political Disinformation Cohere
Medium
Acceptable Use and API Usage Restrictions Cohere
Medium
No Model Training on Enterprise Customer Data Cohere
Medium
Enterprise Customer Data Ownership Cohere
Medium

Showing 30 of 351 provisions. View all →

Official Source

View official regulation text →

Get alerted when platforms change their policies — including -relevant provisions.

Subscribe to Monitor — $19/mo

Frequently Asked Questions

What does require?

Which platforms does apply to?

ConductAtlas tracks -relevant provisions across 56 platforms. Each platform's specific provisions are classified by severity and mapped to requirements.

How does ConductAtlas monitor compliance?

ConductAtlas captures policy documents daily, classifies provisions by regulatory framework, and flags changes that affect obligations. Every change is archived with cryptographic verification.