Definition
Fair use is Fair use is a US copyright defense that lets some uses of protected work happen without permission, judged case by case on four factors, and it is the central question in lawsuits over AI training.
Fair use is a doctrine in US copyright law, written into 17 U.S.C. § 107. It says some uses of copyrighted material do not count as infringement, even without the owner's permission. Courts decide each case on its facts, which is why nobody can say in advance whether a given use is safe.
A court weighs four things:
No single factor decides the outcome. Fair use is also a defense. The defendant has to raise it after the owner says infringement happened.
Models are trained on huge amounts of training data, much of it copyrighted. AI companies argue that training is transformative, because the model learns patterns and does not store a copy for readers. Rights holders argue that the copying is wholesale and that the outputs compete with the originals in their own market.
The US Copyright Office examined this in a 2025 pre-publication report on generative AI training. It said the analysis is highly fact-specific, that some uses will qualify as fair use and others will not, and that it focused mainly on the first and fourth factors. Commentators disagree about how much that favors either side.
Fair use is not a license, and it is not a safe harbor you can claim up front. It is an argument you make if you are sued. The same training run might be fair use for one set of facts and not for another, for example if the model can reproduce a work closely.
Output matters too. A model that produces something substantially similar to a protected work faces a different question than one that only learned from it. Claims about a person's voice or likeness are usually separate legal theories, such as the right of publicity.
Fair use sits next to AI data licensing, where companies pay to avoid the question, and web scraping, the usual way data is collected. Courts are still working out how the four factors apply to large language models and other generative systems.
There is no single answer. US courts decide case by case on four factors, and the Copyright Office has said some AI training uses will qualify and others will not.
No. A license is permission from the owner. Fair use is a legal defense you argue after someone says you infringed.
Anthropic published a revised Usage Policy on Thursday, effective November 12, that bans building surveillance tools, extends the weapons ban to arming drones and adds a rule against cruelty to its models. The cruelty rule got the headlines. The surveillance and policing rules are the ones that can end a product.
OpenAI told investors its annualized revenue is approaching $50 billion, the Financial Times reported, per TechCrunch on Thursday, about $20 billion below the figure that circulated a week earlier. The gap comes from how the number was counted, and it makes every AI run-rate comparison suspect.
Vercel's October AI Gateway index says open-weight models handled more than seven in ten tokens in September and nearly one in four dollars, up from one in twenty in June. Cheap models do the volume now, and nobody is loyal.
Anthropic opened a US waitlist for Claude Duet, a research preview agent on Opus 5.5 that sends a morning brief and works in the background for Pro and Max users. The catch is that it spends the same usage limits you code with.
Meta says a Claude tool reading profiles, posts and hashtags at about four times the normal rate tripped its bot defenses and got the No Kings Coalition's Instagram account suspended Saturday. If your agent browses while logged in as you, Meta judges you by its pace.