Anthropic Faces Copyright Infringement Charges in AI Training Case
Anthropic has been charged with copyright infringement in a California federal court case brought by music publishers; the charges are allegations, and Anthropic is presumed innocent unless proven guilty.
3 reports on this incident · first at Aug 31, 2026, 9:22 p.m. ET
Earlier reports
Aug 31, 2026, 9:22 p.m. ET
Anthropic Charged with Copyright Infringement in AI Training Case
A California federal court has charged Anthropic with copyright infringement, according to a report from wsau.com. The charges, announced on August 31, 2026, come in a case brought by music publishers who alleged that Anthropic pirated hundreds of song lyrics and sheet music from artists including The Beatles, Taylor Swift, and Michael Jackson to train its Claude AI models. The publishers also alleged that Claude can reproduce copyrighted lyrics verbatim when prompted.
The matter is at the charging stage and has not been decided. Anthropic has not been tried or convicted, and is presumed innocent unless proven guilty. The case remains pending in the same California federal court.
Aug 31, 2026, 9:22 p.m. ETFirst report
Music Publishers Sue Anthropic Over Alleged Copyright Infringement in AI Training
Music publishers filed a lawsuit against Anthropic in California federal court on August 31, 2026, alleging copyright infringement. The publishers claim Anthropic pirated hundreds of song lyrics and sheet music from artists including The Beatles, Taylor Swift, and Michael Jackson to train its Claude AI models, and that Claude can reproduce copyrighted lyrics verbatim when prompted.
The case is at the charging stage, and the allegations have not been tested in court. No trial has been held, and Anthropic is presumed innocent unless proven otherwise. Anthropic has not yet responded to the claims.
Was this report accurate and useful?
Sources
Revision history
- Version 11 Sept 2026, 01:22current
First published.
- Version 11 Sept 2026, 01:22current
First published.
- Version 11 Sept 2026, 01:52current
First published.
How we work
This site models an investigative reporter rather than a wire desk. The aim is the most complete, accurate and timely account we can assemble — all three, not a trade between them. Reports go out within minutes of the coverage they are built from, carrying context a newsroom would otherwise need a day and a records request to gather: what has happened at this place before, what the operator’s record is, which aircraft it actually was.
Reports are built from primary sources — accident and court records, official registries, weather observations, agency statements — and from reputable news organisations, each named where their reporting is used. Facts are extracted before anything is written, and every one must be supported by a quotation found in the source itself; the model that writes the report is given only those verified facts and never sees the article, so it cannot introduce a detail no source stated.
Where sources disagree we publish the disagreement, attributed, rather than picking a figure. Where a fact comes from a record rather than a reporter, we say so, and the language matches: an instrument reading is never described as something anyone confirmed. Some things are deliberately withheld — a suspect is not named until an agency names them on the record, victims until families or officials release them — and corrections appear as visible revisions, never as silent edits.
None of that makes a report true. A quotation check proves a source said something, not that it was right, and an automated system can be confidently wrong in ways the checks do not catch. If something here is wrong, the feedback above is how it gets found. The full methodology, including what we refuse to publish.