Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

For U.S. publishers, the key point is that AI training with copyrighted works raises legal and policy questions; it is not automatically lawful or automatically license-required in every case. The U.S. Copyright Office’s Copyright and Artificial Intelligence, Part 3: Generative AI Training, released in pre-publication form on May 9, 2025, examines those questions and discusses voluntary licensing as one possible route. It does not decide whether a particular use is lawful or establish that licensing will work for every publisher, work, or AI developer.

What the Copyright Office report does—and does not—say

The Office’s Part 3 report addresses generative AI training on copyrighted works, potential liability, and licensing. It is agency analysis, not a binding rule that settles every dispute. Whether a specific training use requires permission depends on the facts and applicable law; legal developments after the report may also matter.

This discussion is U.S.-focused. The report should not be treated as a statement of copyright rules in the EU, UK, or any other jurisdiction. Nor does it resolve a publisher’s individual contracts, ownership records, or rights chain.

Keep three questions separate

  • Legal: Does this particular use of this work require permission under the law that applies?
  • Commercial: Can the parties agree on a license, and does its scope fit the intended training use?
  • Operational: Can the publisher identify the relevant rights, show that it controls them, and document what it has authorized?

The Office’s analysis discusses licensing approaches and policy issues; it does not answer these questions for an individual publisher or a particular dataset.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
#1 Best Overall

How voluntary AI-training licenses can be arranged

The report discusses two broad routes: a publisher or other rights holder can negotiate directly with an AI developer, or rights holders can authorize a collective organization to administer licensing on their behalf. The Copyright Clearance Center is cited as an example of collective licensing infrastructure for textual works. That example is descriptive, not a recommendation to join or buy a particular service.

Route How it works Questions for a publisher
Direct license The rights holder and AI developer negotiate with each other. Can the publisher identify and control the rights needed? Does the agreement cover the intended training use? The Office does not quantify transaction costs or prescribe contract terms.
Collective license An organization authorized by multiple rights holders can aggregate rights and administer licensing. The Office names the Copyright Clearance Center as an example for textual works. Does the organization have authority over the relevant works and uses? Does its license scope fit the intended use? The Office does not establish that a collective license covers every publisher or work.

These are possible licensing mechanisms, not proof that permission is legally required in every case or that a license will be available for every category of work. The report also identifies open questions about whether voluntary licensing is feasible and sufficient to meet the needs of AI developers.

What publishers should check before discussing a license

A licensing conversation is only useful if the parties can establish what rights are available and what the proposed permission covers. The Office discusses licensing options, but it does not determine the rights held under any particular publishing agreement.

  • Identify the works. Establish which texts or other works are in scope and whether the publisher can connect those works to its records.
  • Check authority. Review whether the publisher controls the rights relevant to the proposed use, rather than assuming publication alone establishes that authority.
  • Define the use. Ask whether the proposed license covers the training activity and scope the parties intend to authorize.
  • Consider administration. Compare direct negotiation with collective administration in light of the rights to be licensed and the work of identifying and documenting them. The Office does not provide a quantified cost or burden comparison.
  • Keep the legal analysis distinct. A negotiated license is a commercial arrangement; its availability does not itself resolve whether an unlicensed use would be lawful.

What is known about licensing activity—and what is not

The Copyright Office reported that commenters described licensing agreements in several sectors as of the end of 2023, with additional activity afterward. This is qualitative evidence that agreements had been reported; it is not a complete market census, a count of agreements, or proof that licensing is feasible for every type of work.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

The report does not establish a numeric estimate of the AI-training licensing market or settle whether voluntary deals can meet the needs of all parties. Publishers should therefore treat the existence of some reported agreements as evidence of activity, not as evidence of universal availability or adequacy.

Why the Office’s press-publisher study is a separate issue

In its June 30, 2022 Study on Ancillary Copyright Protections for Publishers, the Office concluded that press publishers had significant protection under existing U.S. law and did not recommend a new ancillary copyright. It also found that concerns about journalism funding did not appear to be copyright-specific.

That was a U.S. policy conclusion about proposals for a new ancillary right for publishers. It does not answer whether a particular AI system’s training use is lawful, and it should not be generalized to other countries.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

How to read the scale of the AI inquiry

The Copyright Office says it received more than 10,000 comments in its AI inquiry by December 2023. That figure counts comments, not unique viewpoints, public opinion, training-data volume, or licensing deals. It indicates the inquiry drew a large volume of submissions, but it does not measure consensus on the legal or policy questions.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.