New York Times Sues Brave Over AI Content Reproduction and Paywall Bypass

On November 2, 2026, The New York Times Company filed a federal lawsuit against Brave Software. The publishers of the New York Daily News, Chicago Tribune, and Orlando Sentinel filed a separate complaint. They allege that Brave copied and distributed their journalism through its search index, AI products, and commercial API, and that a browser setting circumvented subscription restrictions. These are allegations, not findings of liability. Read the Times complaint ↗ and the three publishers' complaint ↗.

One Times example is especially striking. Its complaint says a response from Brave's LLM Context API reproduced 1,291 of 1,295 words of an article, about 99.7% of the original. The filing also describes an Ask Brave interaction that allegedly supplied more than 400 verbatim words from a paywalled article after a follow-up request. Those examples warrant scrutiny, but selected examples do not establish how often Brave's products behave that way. I have not independently replicated them.

The lawsuits put a practical question in front of AI search providers: when does a tool help readers find and understand reporting, and when does its output replace the reason to visit the publisher?

What the lawsuits allege

The complaints raise distinct issues. They allege copying into a search index and delivery of article text through AI answers or an API. They separately challenge a browser setting that, the publishers say, permits access to certain subscription pages without logging in. They also make advertising-related claims. Each claim will have its own factual and legal questions; ordinary ad blocking should not be treated as proof that an AI answer reproduced an article.

Brave describes its LLM Context API ↗ as returning extracted page chunks and source metadata for AI applications. That description explains why an API response can carry more than a conventional search snippet. It does not establish that Brave returned any particular newspaper article in the manner alleged. The complaints supply those examples, and the court has not resolved them.

What the examples show, and what they do not

The Times' 1,291-of-1,295-word example is a measure of a response alleged in a complaint, not an independently measured rate of reproduction across the service. The Ask Brave example concerns a different product and interaction. The paywall allegation concerns access controls, not just answer length. Collapsing all three into one claim would obscure where a safeguard might work and where it might not.

A product could check the amount of verbatim overlap in one answer, the share of a source reproduced, and whether repeated requests collectively reconstruct most of a work. A short answer can still contain a valuable part of a story, so no fixed word count or percentage automatically settles fair use. The U.S. Copyright Office ↗ describes fair use as a case-specific analysis. Output limits also would not answer a separate allegation about circumvention of a paywall.

Where I draw the line

I use AI to compare evidence, find contradictions, and examine how a story connects to other sources. That work can create analysis of value to readers. It can also send readers to the original reporting. Whether it actually increases publisher traffic or subscriptions needs measurement, not assumption.

Returning nearly all of an article through an AI product or commercial API, if proven, is different from linking to it, criticizing it, or analyzing its claims. The value of an intelligence platform should come from the intelligence it creates, not from redistributing a publisher's protected reporting. Copyright generally protects original expression rather than facts or ideas, but that principle does not grant a blanket right to reproduce an article.

Could this have been prevented?

An output check is one layer. Source provenance is another. A crawler may encounter a copy of an article on an unrelated site and have no reliable indication whether it is authorized syndication or an unauthorized mirror. The place where text was found is not necessarily the party that may license its reuse. That possibility does not establish what happened in these lawsuits. It shows why AI systems need to distinguish where material was retrieved from who can authorize its use.

Providers can identify canonical publishers, heed access metadata, limit verbatim output, and link prominently to the original. Those steps can reduce risk. They are not a substitute for permission where permission is required, and they do not decide the cases against Brave.

A proposal: the News Discovery Brief

What if publishers supplied a small, public, machine-readable brief that AI search services could use to help readers discover an article without reproducing its protected body? I call this proposed format a News Discovery Brief, or NDB. It is an idea for discussion, not an adopted standard or a product feature already deployed by AINewsIQ.

An NDB could contain the canonical article URL, publisher and author, publication and update dates, access status, a short publisher-approved public brief, and explicit terms for using that brief. The publisher could add who, what, when, where, why, and how when those facts are suitable for a preview. An investigation should not have to reveal its central finding just to become discoverable.

Here is an illustrative excerpt, not a complete schema or an enforceable license:

{
  "format": "ndb/0.1-proposal",
  "canonical_url": "<publisher article URL>",
  "access": "subscription",
  "public_brief": "A publisher-approved preview of the reporting.",
  "brief_terms": "See publisher terms for permitted use"
}

The proposed flow is simple: find the canonical source, read a public discovery brief if the publisher offers one, show that brief with a clear link to the article, and keep independent analysis tied to evidence that can legitimately be used. If no trustworthy brief exists, a service can fall back to conservative metadata and a link rather than assuming a public mirror authorizes reproduction.

This would build on existing tools. Schema.org's NewsArticle ↗ describes article metadata. Google's paywalled-content guidance ↗ uses `isAccessibleForFree: false`, and its preview controls ↗ include `max-snippet`. IPTC NewsML-G2 ↗ addresses news metadata and rights information. NDB's proposed addition is a compact, publisher-approved discovery payload for AI search. It would not override those standards, force crawlers to comply, or prevent lawful independent criticism.

The test that matters

The proposal should be evaluated rather than declared a solution. Compare conventional snippets, publisher-written discovery briefs, linked AI analysis, and comprehensive AI summaries. Measure source attribution, visits to the original, reader understanding, subscription behavior where observable, and how often an answer substitutes for reading the article. Different kinds of journalism may produce different results.

The Brave complaints show why the distinction matters now. Their specific allegations still need to be tested in court. In the meantime, publishers and AI providers can test a more constructive design: give readers enough to understand why a story matters and a clear path to the reporting itself, while preserving room for genuine analysis.