On November 2, 2026, The New York Times (NYT) sued Brave Software, the developer of the Brave web browser, in the U.S. District Court for the Southern District of New York, alleging copyright infringement among other claims. The same day, three other newspapers, the Daily News, the Chicago Tribune, and the Orlando Sentinel, filed a separate complaint. Brave has argued until now that a search engine's indexing of articles and its display of short excerpts (snippets) and summaries qualify as fair use. In this lawsuit, however, the key issue is likely to be not only the act of indexing articles but also how much of an article's full text Brave supplied through its search API. According to the NYT complaint, the "LLM Context" endpoint of the Brave Search API returned 1,291 of the 1,295 words in one article, or 99.7%.
The Copyright Question Centers on How Much Article Text the Search API Provides
The NYT's complaint brings five claims. Three of them concern copyright infringement: direct infringement, on the theory that Brave itself copied the articles; contributory infringement, for facilitating infringement by customers; and vicarious liability, for profiting from infringing activity that Brave was in a position to control. On direct infringement, the complaint takes issue not only with copying made when articles are ingested into a search index or database, but also with providing the content of articles to third parties. The complaint also includes a claim that circumventing paywalls (restrictions on access to paid articles) violates Section 1201 of the Digital Millennium Copyright Act (DMCA), as well as a claim for restitution of unjust enrichment based on revenue Brave allegedly earned by swapping out ads. The NYT is seeking damages, disgorgement of profits, and a permanent injunction. However, paywall circumvention through the "Paywall blocker" and the ad replacement raise legal issues separate from the amount of article text the API provides.
The search API's output matters because the NYT has long objected to the provision of content that substitutes for its articles. According to the complaint, the NYT said in a letter sent in April 2026 that Brave was displaying content that stands in for articles directly to users, and that this could not be justified as fair use. That argument concerns Brave's services for general users, but the same logic could apply to a search API that hands article text to third parties.
Brave runs its own search engine and has built a search index by crawling web pages. In a February 2026 blog post, the company said the index had reached 40 billion pages. According to the NYT's complaint, Brave shifted in 2023 to monetizing its search index and moved forward with the Brave Search API, which supplies search data to AI companies and others. Citing Brave's own materials, the complaint says the API's main uses are training and grounding several major large language models (LLMs). Grounding is a technique in which an AI refers to external search results or documents when generating an answer and uses their content as the basis for that answer.
In the same blog post, Brave also said the Search API handles billions of calls per week. The NYT cites that figure and argues that API use by AI companies far exceeds use by individual users. Against this backdrop of large-scale use, the scope of the information the search API returns has emerged as a copyright problem.
The most notable item in the complaint is the example of article text provided through the LLM Context endpoint. According to the NYT, the endpoint returned 1,291 words verbatim from a 1,295-word article, meaning 99.7% of the original text could be retrieved. In another test, using the title of a different article across three endpoints, LLM Context reportedly returned nearly half the article and Web Search more than a third. The amount of information differs greatly between the traditional approach of attaching a few words of excerpt to a search result and an approach that gives an AI between half and nearly all of an article's text as reference material, even though both come from the same search service.
What Brave has defended in its earlier litigation with News Corp is mainly the building of a search index and the display of snippets and rough summaries. According to Reuters, Brave argues that such conduct qualifies as fair use. News Corp, for its part, filed a counterclaim on July 21, 2026, in the U.S. District Court for the Northern District of California, seeking an injunction and damages of up to $150,000 per infringement. The NYT had also cited the same figure in an earlier warning letter. According to the current complaint, the NYT asked Brave CEO Brendan Eich to stop the conduct at issue in August 2025 and April 2026, but received no response from Brave.
Does the 2015 Ruling Allowing Indexing Also Cover Large-Scale Output for AI?
A potentially important precedent for Brave is the 2015 ruling in the Google Books case. The U.S. Court of Appeals for the Second Circuit held that Google's scanning and digitizing of books to build a searchable database was fair use. The Southern District of New York, where this suit was filed, sits within the Second Circuit, so that decision is important for assessing Brave's indexing. If the dispute were limited to collecting and indexing articles, that could work against the NYT.
However, what the Google Books ruling permitted was digitizing books to make them searchable and offering snippets with limited display. Whether that reasoning extends to handing nearly the entire text of an article to a third party's AI system is a separate question. The NYT's 99.7% figure highlights precisely this difference. Even if indexing articles is permissible, whether a state in which nearly the full text can be retrieved from a search API qualifies as fair use will need to be decided anew.
That said, caution is warranted in generalizing the figures in the complaint. The 99.7% figure comes from one specific article combined with the LLM Context endpoint, and it does not show that the same proportion of text is returned for every article. In the test using another article, LLM Context reportedly returned nearly half and Web Search more than a third, so the share of text that can be retrieved varies. Moreover, all of this reflects the plaintiffs' claims, and no rebuttal from Brave had been confirmed at the time of reporting.
The complaint names Anthropic, Perplexity, Mistral, xAI, and Cohere as customers of the Brave Search API. However, these AI companies are not defendants in this lawsuit. A footnote in the complaint also points out that Brave's API introduction page and company information pages list Cohere, Mistral, Perplexity, and You.com as users. The NYT sued not the AI companies that use the API, but Brave itself, which provides the search index and the API.
What deserves attention in this case is whether the fair-use debate over traditional search engines applies unchanged to supplying information to AI. The display of snippets and summaries that Brave has defended and the provision of nearly full article text alleged in the complaint may differ in how much users need to access the original work. On that basis, the NYT can be read as arguing that Brave's legitimacy as a search service alone cannot justify the API's large-scale provision of article text. This reading, however, assumes the output described in the complaint is factual, and the assessment could change depending on Brave's explanation and future evidence.
Brave's Response Will Be Key, and There Are Parallels with the Perplexity Suits in Japan
Two conditions will largely shape what happens next. First, Brave's answer or submitted evidence may show that what LLM Context returns is in fact limited to a portion of an article. If the API specifications set a clear cap on the number of words it can return, and the large-scale text retrieval described in the complaint cannot be reproduced, the argument centered on output volume would weaken. Second, a court could decide that the provision of large amounts of text qualifies as fair use. As far as could be confirmed, including in the News Corp litigation, no court has yet ruled on these issues.
In Japan, too, lawsuits are under way over AI's use of news articles. The Asahi Shimbun Company and Nikkei Inc. sued Perplexity in the Tokyo District Court over unauthorized use of their articles, seeking an injunction and a total of 4.4 billion yen in damages (2.2 billion yen each), and the first oral argument was held on May 14, 2026. Three Yomiuri Shimbun companies are also seeking about 2.1 billion yen in damages from Perplexity, with their first hearing held on May 26 of the same year. Perplexity has asked for the claims to be dismissed in all of these cases.
The NYT's complaint names Perplexity as a customer of the Brave Search API. But the party whose conduct is at issue differs between the Japanese suits and this U.S. suit. The Japanese cases question the generation of AI answers and the copying involved in using articles, whereas the NYT is pursuing the responsibility of Brave, which builds the search index and supplies information through the API. In other words, the suits target different stages of how AI uses information. Nor can it be confirmed from this complaint whether the Brave Search API is used to generate AI answers for Japan.
The next important points of reference will be the answer Brave files and the actual specifications of LLM Context. Once it becomes clear how many words of article text a single API call can return, and whether the output examples in the complaint can be reproduced, both sides' claims can be evaluated more concretely. That should show whether the dispute narrows to where to draw the line on how much text is retrieved and provided, or widens to the responsibility of companies that supply news articles to AI in general.