CLEAR Act
- Sponsor
- Sen. Schiff, Adam B. [D-CA]
- Committees
- Judiciary Committee (primary)
- Last action
- Feb 10, 2026
Bottom line
The CLEAR Act establishes a new federal reporting requirement for generative AI developers regarding their use of registered copyrighted materials in training datasets, aiming to enhance transparency and provide enforcement mechanisms for copyright holders.
What it actually does
This bill, known as the CLEAR Act, mandates that any person using a training dataset containing copyrighted works to develop or release a generative AI model must submit a detailed notice to the Register of Copyrights. The notice must summarize each copyrighted work used and include its URL if publicly available. Non-compliance can result in civil penalties, injunctions, and attorney's fees, with penalties directed to the Copyright Office. The bill also requires the Register to establish a publicly accessible online database of these notices.
Proponents argue
Proponents argue this bill provides essential transparency for copyright holders, allowing them to identify when their registered works are used in AI training datasets. This transparency is crucial for protecting intellectual property rights in the rapidly evolving AI landscape and ensuring fair compensation or control over their creative works. It establishes a clear mechanism for accountability without stifling innovation, creating a public record that can inform future licensing and fair use discussions.
Opponents contend
Opponents might argue that the reporting requirement is overly burdensome for AI developers, potentially slowing innovation and creating a massive, unmanageable database. They may contend that identifying and summarizing every registered copyrighted work in vast training datasets is technically challenging and that the definition of 'copyrighted work' (registered or scheduled) is either too narrow or too broad. Concerns could also be raised about the potential for frivolous lawsuits and the impact on open-source AI development.
The bill is short and relatively straightforward, making it accessible for an informed reader to evaluate its core provisions within a reasonable timeframe.
Section 2(a)
Definitions for AI, Copyrighted Work, Generative AI Model, and Training Dataset
This section defines key terms used throughout the bill. 'Artificial intelligence' is broadly defined as automated systems performing human-like cognitive tasks. A 'copyrighted work' is specifically limited to works protected and *registered* under Title 17. A 'generative AI model' is defined as computer code and numerical values designed to create expressive material using AI. A 'training dataset' is a collection of material units used for training.
Supporters argue
Proponents argue that clear and precise definitions are essential for legal certainty, ensuring that the bill's requirements are understood and applied consistently by both AI developers and copyright holders. The specific definition of 'copyrighted work' to include only registered works focuses the bill on easily identifiable intellectual property, making compliance more practical.
Critics contend
Opponents might contend that the definition of 'copyrighted work' as *registered* works is too narrow for some copyright holders who believe all protected works should be covered, or that the broad definition of 'artificial intelligence' could encompass systems not intended by the bill's primary focus, leading to unintended compliance burdens.
Tradeoffs
This provision balances the need for clear legal definitions in a complex and rapidly evolving technological and legal landscape with the practicalities of implementation. Limiting 'copyrighted work' to registered items simplifies compliance but excludes a large body of creative work from the bill's direct transparency mechanisms.
Section 2(b)
Requirement to Submit Notice to Register of Copyrights
This provision mandates that anyone using a training dataset containing copyrighted works to train or release a generative AI model must submit a notice to the Register of Copyrights. This notice must include a sufficiently detailed summary of each copyrighted work in the dataset and, if publicly available, its Uniform Resource Locator (URL). The notice must be filed 30 days before commercial use or release for new models, or 30 days after regulations are issued for existing models.
Section 2(c)
Enforcement and Penalties for Non-Compliance
This section establishes a cause of action for copyright owners if their registered work is used in an AI training dataset without the required notice being filed. Courts can impose civil penalties of at least $5,000 for each instance of failure to submit notice, with these funds paid to the Register of Copyrights to offset operating costs. Courts can also issue injunctions to halt the use of the applicable work until notice is filed, and award attorney's fees and expenses. An annual cap of $2,500,000 is placed on total civil penalties per person, but injunctions and attorney's fees are not subject to this cap.
Section 2(d)
Establishment of Publicly Available Online Database
This provision directs the Register of Copyrights to establish and maintain an online database that makes all submitted notices publicly available. This database will serve as a central repository for information regarding copyrighted works used in training generative AI models.
Defines 'copyrighted work' as a work that is protected under title 17 AND *registered under section 408, or scheduled pursuant to section 1401, of title 17, United States Code*.
Section 2(a)(2)(B)
Why it matters:This is standard legislative practice for defining the scope of a bill, but its impact is profound. It significantly narrows the applicability of the bill compared to covering all works protected by copyright, which do not necessarily need to be registered to receive protection.
Case for: Proponents would argue that limiting the scope to registered works makes the notice requirement practical and enforceable. Registered works have a clear public record, making it easier for AI developers to identify them and for copyright holders to prove ownership and use. It avoids the immense burden of identifying every single copyrighted work, registered or not, in a dataset.
Case against: Critics, particularly some copyright holders, might argue that this limitation leaves a vast amount of copyrighted material unprotected by the bill's transparency requirements. Many works are protected by copyright upon creation but are never formally registered, meaning their use in AI training would not trigger the notice requirement, creating a significant loophole.
Estimated impact: This provision significantly limits the universe of works for which notice is required, potentially reducing the compliance burden on AI developers but also reducing the transparency for a large segment of copyrighted content that is not formally registered.
Requires the notice to contain 'a sufficiently detailed summary of each copyrighted work in the training dataset.'
Section 2(b)(1)(A)
Why it matters:This is common legislative practice to delegate specifics to agency rulemaking, allowing for flexibility and adaptation. However, it means the true burden and utility of the notice requirement are not yet known, as they depend entirely on how the Register of Copyrights defines 'sufficiently detailed.'
Case for: Proponents would argue that this flexibility allows the Register of Copyrights, as the expert agency, to develop practical and effective guidelines that can adapt to technological changes and industry best practices. It enables a nuanced approach rather than a rigid, potentially outdated, statutory definition.
Case against: Opponents would contend that this ambiguity creates significant uncertainty for AI developers, who won't know the full extent of their compliance obligations until regulations are issued. This could lead to over-compliance, under-compliance, or significant legal disputes over what constitutes 'sufficiently detailed,' creating a chilling effect on innovation.
Estimated impact: The ultimate effectiveness, practicality, and burden of the notice requirement will heavily depend on the regulations issued by the Register of Copyrights defining 'sufficiently detailed summary.' This definition will be a critical determinant of the bill's real-world impact.