California AB 2013 — Generative Artificial Intelligence: Training Data Transparency

California Assembly Bill 2013 establishes the first U.S. state law requiring transparency in generative-ai training data. The law, effective January 1, 2026, mandates that ai-system-developers must publicly disclose detailed documentation about datasets used to train generative AI systems made available to Californians.

The law requires comprehensive disclosure including dataset sources and owners, data point counts, types of data, intellectual property status, personal information inclusion, data cleaning processes, collection timeframes, and use of synthetic-data-generation. Developers must provide this documentation before making systems available and before each substantial modification.

Key exemptions include systems solely for security and integrity purposes, aircraft operation in national airspace, and national security applications limited to federal entities. The law applies to systems released on or after January 1, 2022, covering both free and paid services.

This legislation represents California's approach to AI transparency, complementing the state's broader AI governance framework alongside colorado-ai-act-sb24-205 as pioneering state-level AI regulation in the United States.