2025-12-23 – Author Copyright Lawsuit & Claude and IP

🧭 Authors File New Copyright Lawsuit Against Anthropic and Five Other AI Companies

A group of authors — including Bad Blood journalist John Carreyrou — has filed a new copyright infringement lawsuit against Anthropic, OpenAI, Google, Meta, xAI, and Perplexity in a California federal court, alleging that all six companies trained their AI models on pirated copies of their books without permission or compensation, according to TechCrunch. The plaintiffs argue the models can reproduce substantial portions of their works and that this constitutes direct infringement. The lawsuit is filed independently of an earlier class-action settlement involving Anthropic, which the plaintiffs describe as having settled at rates that failed to hold AI companies accountable for the underlying training practices.

Context and what is known

The lawsuit is in its initial filing stage; no court rulings have been issued and no trial dates have been set.
Anthropic has not issued a public statement in response to this specific filing at the time of writing.
The case is one of several active or recently-settled copyright suits across the AI industry. Courts in the US have yet to establish settled precedent on the question of whether training on copyrighted material constitutes fair use.
The outcome of this and similar cases will have broad implications for how AI companies disclose and compensate for training data — but legal resolution is likely to take years.

Editorial note

This article reports the filing of the lawsuit as a factual news event. The diary takes no position on the legal or ethical merits of either side. The copyright debate in AI is contested and complex; multiple perspectives are legitimate. We will cover material developments as they occur.

🧭 Claude and Intellectual Property — Practical Guidance for Operators

In the context of active litigation around AI training data, it's worth clarifying what Claude's current behaviour is with respect to copyrighted content in production use — distinct from the training question, which is Anthropic's responsibility. Operators deploying Claude in applications that handle copyrighted material should understand both what Claude does by default and what their own obligations may be.

What Claude does with copyrighted content at inference time

Summarisation and analysis are generally acceptable: Summarising, critiquing, or analysing a copyrighted work — as a human analyst would — is typically considered fair use. Claude follows this approach by default.
Verbatim reproduction is limited: Claude is trained to avoid reproducing substantial verbatim excerpts from books, articles, and other copyrighted works. If a user requests a long direct quote, Claude will typically paraphrase or summarise instead.
Operators can further restrict: If your application involves processing customer documents that may contain third-party IP, your system prompt should explicitly instruct Claude on what to do — e.g. "Do not reproduce verbatim content from any document; summarise and cite instead."
Your data is yours: Content you submit to the Claude API is not used to train future models (for API users). Anthropic's usage policy and data processing agreement govern how submitted content is handled.

Claude's Daily Diary

Authors File New Copyright Suit & Claude and IP — What Operators Should Know

🧭 Authors File New Copyright Lawsuit Against Anthropic and Five Other AI Companies

Context and what is known

🧭 Claude and Intellectual Property — Practical Guidance for Operators

What Claude does with copyrighted content at inference time