Congress Weighs Transparency Rules for AI

Judge writing at desk with gavel and Lady Justice statue
Photo: sebra / Shutterstock

A federal court said training an artificial intelligence model on copied books can be fair use, and Congress is now weighing a bill that could force companies to reveal what they scraped.

Story Snapshot

  • A judge ruled Anthropic’s training on books was fair use at summary judgment.
  • The ruling covered copying for training and alleged “memorization” during training.
  • The CLEAR Act would require public disclosure of copyrighted works used in training.
  • Publishers and creators say unlicensed training steals work and harms markets.

What The Court Actually Decided In The Anthropic Case

A federal district court in California held that Anthropic’s use of copied books to train a large language model was fair use. The court granted summary judgment for Anthropic on training use, which means the judge found no need for a jury on that point based on the record. The decision focused on training as the legally relevant use, not general publication of the books. The Copyright Office’s summary places this holding at 787 F. Supp. 3d 1007 (2025).

The court’s reasoning reportedly covered copying done for training and any claimed “memorization” that happens during that process. That matters because many creators fear models store and spit back their lines. The World Intellectual Property Organization’s summary states the court treated alleged memorization as part of the training fair use analysis. This narrows one core dispute to what the model does in production, not what it saw in training.

Why This Is Not A Blanket Green Light

The Anthropic ruling is powerful for that case, but it is still a district court decision. It does not set a nationwide rule for all models, datasets, or business uses. Other courts could read the facts or the four fair use factors differently. Even in the United States, fair use turns on details like what was copied, how it was obtained, how the model uses it, and whether outputs replace the source market. The government docket shows the order but not a universal standard.

Ongoing lawsuits keep the fight alive. Publishers, including The New York Times, accuse major developers of copying millions of articles and damaging the market for news. Lawmakers have cited those claims to argue that the government should not give artificial intelligence firms a free pass. A House Democratic release framed the issue as “unlimited, free use” of journalism if courts accept broad defenses. These are assertions, not final court rulings, but they shape public debate.

Congress Weighs Disclosure Through The CLEAR Act

The Copyright Labeling and Ethical Artificial Intelligence Reporting Act would force developers to file notices that identify copyrighted works used for training. The text requires a sufficiently detailed summary of each work and, when used, the training dataset’s web address. Supporters say this shines light on who used what and helps resolve claims. The bill was introduced in the Senate during the current Congress controlled by Republicans in both chambers.

Industry groups for publishers and authors back stricter rules. The Association of American Publishers argues that fair use cannot excuse wholesale, unauthorized copying to train generative models. The group also presses for record-keeping and disclosure so rights holders can find who owes payment. Writer and talent advocates add that licensing markets already exist and that unlicensed training undercuts those deals. These positions aim to direct artificial intelligence growth through contracts, not compelled tolerance.

What This Means For Readers, Creators, And Builders

For everyday readers, the split is clear. One court says training can be fair use on the facts it saw. Creators say their work is being taken without consent or pay, and they want disclosure, control, and money. Congress is weighing a transparency mandate that could expose which datasets include which works. The next steps will come from more court rulings and whether Congress forces daylight on training sources and methods.

For both left and right, the risk is the same old pattern. A fast-moving industry races ahead while government trails behind. Agencies and courts argue over rules after the market is set. Citizens see insiders win and paperwork pile up for everyone else. A balanced path would protect innovation that helps workers and small firms, while making sure creators are not left holding the bag. That balance will hinge on facts, not slogans, in the months ahead.

Sources:

foxnews.com, congress.gov, schiff.senate.gov, govinfo.gov, nadler.house.gov

© unitedfrontnews.com 2026. All rights reserved.