Five AI companies are named in The New York Times' new lawsuit against Brave, and none of them is a defendant (PPC Land). Anthropic, Perplexity, Mistral, xAI and Cohere appear as customers of the product the Times went after when it sued in federal court in Manhattan on November 2 (TechCrunch).

The product is the Brave Search API, a common choice for RAG pipelines and agents that need fresh web results. The complaints call it "an illicit source of unlicensed copyrighted material" (PPC Land).

If your app sends queries to that API, you are in the same seat as the five companies the complaint names: a paying customer of a service whose supply chain is now in court. These are allegations, not findings. Nothing about your integration breaks today. But it is a good week to look at how your search layer is wired.

What the Times filed

There are two suits. The Times filed one. The publishers of the Daily News, the Chicago Tribune and the Orlando Sentinel filed a similar one the same day (TechCrunch). Both sit in the Southern District of New York, as cases 26-cv-9549 and 26-cv-9554, and both demand a jury trial (PPC Land).

Each complaint brings direct, contributory and vicarious copyright infringement, DMCA anti-circumvention claims under § 1201, and unjust enrichment under New York law (PPC Land). The Times describes Brave as a "supplier, dealer, and distributor of stolen copyrighted content" and asks for an injunction and damages (TechCrunch).

The suits reach past the API. They also target a browser feature the Times says gets around its paywall, and what the complaints describe as Brave's practice of blocking publishers' ads while selling its own (TechCrunch). According to Search Engine Journal, the Times says it sent cease-and-desist letters in August 2025 and April 2026 and cites them as evidence of willful infringement. The Times published the complaint itself.

A large magnifying glass made of folded newspaper hovering over a web of crawled pages, with thin cables running from the web to five small anonymous robot silhouettes
The Times' claim, in one picture: pages go into Brave's index, and the index goes out through the API.

Why the Search API sits in the middle

The complaints name four endpoints: News Search, Web Search, LLM Context and Answers. They describe LLM Context as returning "extracted page chunks and source metadata in a shape and length that satisfies the needs of an LLM," and say Answers is "the same service that powers the Ask Brave product" (PPC Land).

A classic search API returns links and short snippets. The Times' argument is that an endpoint built to hand a model page content to reason over is closer to delivering the article itself. Brave's own API page lists LLM Context as "agentic search" and, as of November 2026, prices the Search plan at $5 per 1,000 requests and Answers at $4 per 1,000 requests plus $5 per million input/output tokens.

The crawler is the other half. The Times says Brave built an index of more than 30 billion pages with a crawler it never identified, which publishers cannot block (TechCrunch). Brave's crawler help page says the bot "does not advertise a differentiated user agent," because Brave "must avoid discrimination from websites that allow only Google to crawl them." The same page says Brave won't crawl what Googlebot can't, and that a noindex directive delists a page.

By Brave's own description, then, a publisher that wants Google but not Brave has no separate robots.txt line to write, and the complaints lean on exactly that.

Brave's side

Brave did not immediately respond to TechCrunch's request for comment (TechCrunch). It has argued a close cousin of this case before. Brave sued News Corp first, after a cease-and-desist letter, arguing that indexing website content is something "all search engine operators must do to exist." News Corp filed its own scraping suit against Brave in July 2026 (Semafor).

A search engine does have to copy pages to index them at all. The open question is whether serving extracted page content to AI companies is still "search." No court has answered it in this case. Expect years of motions before anyone gets a verdict.

What to do if your app calls Brave today

I'm not a lawyer, and this isn't legal advice. If your company has real exposure, ask yours. From the engineering side, four things are worth an afternoon.

  1. Read your plan's storage terms. Brave's API FAQ answers "Can I store results or use them to train models?" with "Yes, with a plan that includes storage rights." If you cache search results in a vector store or keep them for evals, check that your plan covers it.
  2. Record where each citation came from. If a grounding source is ever challenged, you want to know which answers it fed.
  3. Keep the provider behind one interface. The complaints ask for a permanent injunction against "the unlawful conduct alleged herein," wording PPC Land reads as broad enough to reach the API endpoints. If Brave is called from twenty places, swapping it is a project. If it is called from one adapter, it is a small change.
  4. Know your fallback before you need it. Pick a second provider now and run your retrieval evals against it, so you already know what a switch costs.

A wall socket shaped like an application module, with three interchangeable plugs labelled only by shape and colour waiting beside it, flat vector
One port, many adapters: the search provider should be the easiest part of your RAG stack to replace.

Points 2 and 3 fit in one small TypeScript interface:

type Hit = { url: string; title: "string; text: string };" interface SearchProvider { readonly name: string; search(query: string): Promise<Hit[]>; } type Citation = Hit & { provider: string; fetchedAt: string }; async function ground( provider: SearchProvider, query: string, ): Promise<Citation[]> { const hits = await provider.search(query); const fetchedAt = new Date().toISOString(); return hits.map((h) => ({ ...h, provider: provider.name, fetchedAt, })); }

Your Brave client becomes one class that implements SearchProvider. Every citation your model sees carries provider and fetchedAt, so once you log them you can answer "which answers used Brave last month" instead of guessing.

Grounding data is the next legal front

The Times has already sued OpenAI and Perplexity on similar claims (TechCrunch). Those cases went after the AI companies themselves. This one goes after a supplier one layer down: the index that feeds models at answer time.

You already track license changes in your npm tree. The source of the text your model quotes back to users deserves the same attention.

Do you know which search provider grounds your app's answers, and how long it would take you to swap it? Tell me in the comments.

If this was useful

I build Hermes IDE, a free, source-available desktop app (macOS, Windows, Linux) for running coding agents like Claude Code and Codex side by side, so weigh my opinions on tooling with that in mind.

Swappable retrieval and evals for your search layer are what the RAG Pocket Guide is about. If this post made you look twice at where your grounding data comes from, it covers retrieval, chunking and reranking patterns for production in a short read.

RAG Pocket Guide: Retrieval, Chunking, and Reranking Patterns for Production