On 2 November 2026, NYT sues Brave in the US District Court for the Southern District of New York, filing a 72-page complaint (case 1:26-cv-09549) that targets an index of "over 30 billion webpages" and the products built on it. Our reading is that this reseller theory is a stronger footing for publishers than the training-use claim that failed at the interim stage in India's ANI v OpenAI case, because it attacks a paid business that sells page text, not the act of training a model. That is an assessment, not legal advice, and the Brave case has not yet been tested by a judge.
TL;DR
- The New York Times Company sued Brave Software on 2 November 2026. Daily News, Chicago Tribune and Orlando Sentinel filed a parallel suit (26-cv-9554) the same day. Both demand a jury trial, per the complaint.
- The target is the data middleman: Brave allegedly copies publisher content into an index and sells access to AI companies, rather than building a frontier model itself.
- Five counts are pleaded, including direct, contributory and vicarious copyright infringement, DMCA circumvention and unjust enrichment.
- Anthropic, Perplexity, Mistral, xAI and Cohere are named as receivers of the content. None is a defendant.
- In India, the Delhi High Court declined an interim injunction against OpenAI on 24 July 2026. The Brave theory covers ground that ruling did not.
Why did NYT sue Brave instead of an AI lab?
Every earlier AI copyright case aimed at a company that trained a model on publisher content or built answers from it. The Brave complaint goes one layer down. It describes Brave as acting "both like a burglar that misappropriates the content of publishers, and an AI 'fence' that sells stolen content to third parties," according to the complaint.
The plaintiffs name the labs as "knowing receivers" but do not sue them. That reads as a deliberate choice to map the supply chain from the reseller upward. The filing also extends an existing sequence: the Times sued OpenAI in December 2023, and the Times and Chicago Tribune sued Perplexity in December 2025, so Brave is the third distinct layer to face suit. The complaint also alleges the index was built by an unidentified crawler that site owners had no way to block.
What exactly is Brave accused of selling?
Three products monetise one asset, the Brave index. The complaint says Brave Search handles more than 50 million queries a day, and about a third lead to an AI-generated response (complaint). The Brave Search API has four endpoints: News Search, Web Search, LLM Context and Answers. The LLM Context endpoint is described, in language the complaint quotes, as page content "ready for LLM consumption" with "no scraping needed."
The filings cite Brave's own 2023 API guide, which says the API's most common uses are training and grounding large language models. They also cite Brave's November 2023 blog post saying Anthropic used the API to train Claude 2, and Meta's Llama 3 paper describing training with Brave's API.
| Brave product | What the complaint says it does | Counts pleaded |
|---|---|---|
| Ask Brave and the Search API | Returns or outputs publisher text, including paywalled articles | Direct, contributory and vicarious copyright infringement (17 U.S.C. § 501) |
| Paywall bypass in the browser | Circumvents access controls | DMCA circumvention (§ 1201) |
| Ad blocking and Brave News | Substitutes Brave's ad arrangements for the publisher's | Unjust enrichment under New York law |
The fifth count, per the complaint, does not depend on any copying. It concerns advertising money, which makes it independent of how the copyright questions come out.
What evidence did the Times include?
The sample outputs are the most concrete part. When given the URL of a paywalled Times article, Ask Brave offered to retrieve the full piece and then output over 400 words verbatim, according to the complaint. Asked for the most reputable newspaper's most prominent article of the day, it produced nearly 700 words of a Times story.
On scale, the Times claims over 10 million registered copyrighted works and more than 13 million paying subscribers as of 30 June 2026, per the complaint and its exhibits. The Tribune exhibits count over 2 million articles across the three papers: more than 500,000 for the Daily News, 1.2 million for the Chicago Tribune and 400,000 for the Orlando Sentinel. TechCrunch reported that Brave had not responded when it covered the filing, and the Times seeks an injunction and damages. The CourtListener docket is the place to follow what happens next.
How does the Brave case compare with India's ANI v OpenAI ruling?
On 24 July 2026, the Delhi High Court declined ANI's interim injunction against OpenAI, treating storage for training as falling under the fair dealing exception in Section 52(1)(a) of the Copyright Act, 1957. The main suit is still pending. The court also weighed the effect an injunction would have on Indian LLM development.
The two cases attack different things. ANI challenged training. The Brave complaint challenges a commercial index that stores publisher pages, serves their text to paying customers and, allegedly, removes paywalls. A fair dealing finding for training does not obviously cover a paid business whose product is the copied text itself. That is why we rate the reseller theory as the stronger one for publishers. Indian publishers, including DNPA members among the OpenAI respondents, have reason to watch this layer.
What does a retrieval price look like next to a licence?
This is our own arithmetic, not a figure from either court. PPC Land reported the Brave Search API priced at $5 per 1,000 requests. At roughly ₹90 to the dollar, 1,000 grounded queries cost about ₹450.
ANI had offered OpenAI a $7.5 million licence in October 2024, per The Hindu. That is about ₹67.5 crore at the same rate, and it equals 1.5 billion API requests at $5 per 1,000. Set against a Times catalogue of over 10 million registered works, the per-request price looks very low next to what a content licence costs. That gap is the economic core of the "fence" argument, though a court will decide the legal question, not the arithmetic.
What should teams building RAG pipelines check now?
If your grounding layer calls a third-party search or LLM-context API, you sit on the receiving side of this theory. Brave is only the named example. A practical checklist:
- List every external index, search API and page-content endpoint in your retrieval path, including those buried inside agent frameworks.
- Ask each vendor in writing where its pages came from, and whether the crawler honours robots.txt and publisher opt-outs.
- Check whether your contract includes an indemnity for third-party copyright claims, and whether it has a cap.
- Separate retrieval of links and snippets from retrieval of full page text, and prefer the former where you can.
- Keep logs of which source supplied which grounding passage, so you can remove a vendor's content quickly if needed.
For the architectural background on grounding, see RAG vs agentic AI in 2026. Teams weighing a self-hosted route can read Perplexica vs Perplexity. On the other legal front over how models learn, see the distillation dispute involving Anthropic and Kimi. For another case of leaked behaviour in a search-linked agent, see the Perplexity coding agent leak. If you are choosing models alongside data sources, this model comparison covers Mistral and Claude.
What are the limits of this analysis?
The complaints are allegations. Brave has not answered in court as far as the sources we reviewed show, and nothing here has been tested by a judge. Brave, founded in 2015 and reported by TechCrunch to have raised roughly $250 million, may argue that its index is a conventional search engine function. Our verdict that the reseller theory is the stronger one is a judgement about the structure of the claims, and a US court could still disagree.
FAQ
Q: Who is suing Brave and where?
A: The New York Times Company filed case 1:26-cv-09549 in the Southern District of New York on 2 November 2026. Daily News, Chicago Tribune Company and Orlando Sentinel Communications Company filed a separate suit, 26-cv-9554, the same day, as the complaint and docket show.
Q: Are Anthropic, Perplexity or other AI companies defendants?
A: No. The complaints name Anthropic, Perplexity, Mistral, xAI and Cohere as receivers of the allegedly copied content, but only Brave Software is sued.
Q: What claims does the New York Times make against Brave?
A: Five counts: direct, contributory and vicarious copyright infringement, DMCA circumvention over the paywall blocker, and unjust enrichment under New York law over ad substitution.
Q: Does the Delhi High Court ANI ruling settle this issue in India?
A: No. It was an interim decision on 24 July 2026 about training and fair dealing under Section 52(1)(a), and the main suit is pending. It did not address a commercial reseller of publisher content.
Q: Does this affect developers who use search APIs for grounding?
A: Potentially, because they are the receiving side of the pleaded theory. Review where your index vendor sources its pages, and check your contractual protections, before the case develops further.
Top comments (0)