社会ガイド
California AB 2013 Training Data Transparency Law
California AB 2013 requires developers of covered generative AI systems and services to publish training-data documentation.
このページでは3 分で読めます
概要
The initial posting deadline was January 1, 2026; developers must also post before each later public release or substantial modification in scope. The law creates transparency duties but does not decide whether particular training uses are lawful.
ディープダイブ
AB 2013, the Generative Artificial Intelligence Training Data Transparency Act, was enacted in 2024. It requires a developer of a covered generative AI system or service made publicly available to Californians for use to post training-data documentation on its website. The initial posting deadline was January 1, 2026. For a covered system released on or after January 1, 2022, documentation must also be posted before each later public release or substantial modification. The documentation includes a high-level summary of training datasets and information about their sources or owners, purpose, approximate number and types of data points, and collection period. It must address whether datasets include copyrighted, trademarked, patented, personal, or aggregate consumer information; whether they were purchased or licensed; how the developer cleaned or processed them; when they were first used; and whether synthetic data generation was used. The definition of developer includes parties that substantially modify a system for public use. AB 2013 requires dataset-level information, not publication of the raw training corpus or source code. A disclosure does not grant copyright permission, decide fair use, or replace privacy obligations. The act includes exceptions, including systems used solely for security and integrity, systems whose sole purpose is aircraft operation in the national airspace, and systems developed for national-security, military, or defense purposes that are made available only to a federal entity. Read each exception against its exact statutory conditions. Maintain a training-data inventory linked to model versions, fine-tuning runs, synthetic-data generation, and release dates. Assign an owner to prepare and update the public documentation, review sensitive claims, and preserve evidence of what was posted for each covered release. Describe uncertainty rather than inventing dataset provenance. Check the enacted text and current California code for definitions, exceptions, and amendments.
戦略的影響
リスクと安全性
AI による壊滅的な被害も日常的な被害も、誰がリスクを理解し、誰が行動できるかにかかっています。
より明確な判決
国民と専門家のリテラシーは、強力な安全政策が政治的に可能かどうかを左右します。
誇大広告を打ち破る
明確な説明は、誇大広告、研究室の PR、曖昧な倫理劇場に囚われることを減らします。
The Future of California AB 2013 Training Data Transparency Law
AB 2013 is operative on a fixed timeline, and implementation choices may be clarified through amendments, enforcement, or litigation. Data pipelines and model versions continue to change, making stale disclosures a practical risk. Developers should monitor official California sources, preserve the dated disclosure in release records, and reassess coverage after substantial model changes. Training-data transparency should be coordinated with privacy, copyright, and contractual review. Retain dated copies of posted documentation and link them to model release records. Recheck the official code after amendments.
現実世界の実装
A developer lists the sources or owners of training datasets and explains how the data were collected, processed, and used.
A model maker describes whether training included copyrighted, trademarked, or patented material, without claiming that disclosure settles permission questions.
A developer identifies whether personal information was included and whether synthetic data generation was used.
A system developed for national-security, military, or defense purposes is checked against the statutory condition that it be made available only to a federal entity.
リスクとガードレール
能力が複雑になる一方で、実存的なリスクを SF として扱います。
高度な自律性の下での調整による表面製品の安全性を混乱させる。
英語以外や専門家ではない聴衆には、低品質の情報源しか提供されません。
実装ロードマップ
製品の危害、誤使用、制御不能/調整不良のリスクを分離します。
どのような証拠がタイムラインと重大度についてのあなたの見方を変えるかを尋ねてください。
マーケティング上の主張よりも、一次情報源と具体的な評価を優先します。
意識だけでなく、キャリア、政策、資金、スキルなど、行動経路を 1 つ特定します。
探検を続けましょう
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the California AB 2013 Training Data Transparency Law quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
よくある質問
What is California AB 2013 Training Data Transparency Law?
California AB 2013 requires developers of covered generative AI systems and services to publish training-data documentation. The initial posting deadline was January 1, 2026; developers must also post before each later public release or substantial modification in scope. The law creates transparency duties but does not decide whether particular training uses are lawful.
Which developer activity can bring a system within AB 2013?
The statute defines developer broadly to include designing, coding, producing, or substantially modifying a system or service for public use.
Which public documentation deadline does AB 2013 set?
The statute sets January 1, 2026 as the deadline for the required posting.
Which information belongs in the required documentation?
AB 2013 calls for documentation about datasets used in development, including source, collection, processing, and use information.
Does AB 2013 disclosure itself authorize use of copyrighted training material?
The transparency requirement does not resolve copyright permission or fair-use questions.
Which system date can be relevant to coverage?
The statute’s scope includes systems released or substantially modified on or after January 1, 2022.
学び続ける
関連ガイド
このトピックのために選ばれたその他のガイド