Basically, the Dictionary Definition of “AI Lawsuit”

Most of the ninety-plus copyright lawsuits filed against AI companies follow a familiar script: a content owner alleges that an AI developer scraped its copyrighted works to train a large language model (“LLM”), and the parties settle in for years of discovery and fair-use briefing. The complaint that Encyclopædia Britannica, Inc. and Merriam-Webster, Inc. filed against OpenAI on March 13, 2026, in the United States District Court for the Southern District of New York, follows that script but adds a chapter that most AI copyright complaints have not attempted. Encyclopædia Britannica, Inc. v. OpenAI, Inc., No. 1:26-cv-02097 (S.D.N.Y. filed Mar. 13, 2026).

In addition to its four copyright counts, the complaint asserts a Lanham Act claim for false designation of origin and trademark dilution. The theory: when ChatGPT generates fabricated content (commonly known as “hallucinations”) and attributes that content to Britannica or Merriam-Webster by name, it misleads users into believing the hallucinated output is endorsed by, or sourced from, two of the most trusted reference brands in the English language. That trademark claim is what makes this complaint worth watching.

The Parties

Britannica, founded in Edinburgh in 1768, is now a global digital education platform that delivers content in over 20 languages to more than 150 million students worldwide. Britannica owns the copyright in nearly 100,000 online articles registered with the U.S. Copyright Office. Compl. ¶¶ 32–37.

Merriam-Webster, a Britannica subsidiary, has been the leading American dictionary publisher for over 180 years. Its print Merriam-Webster’s Collegiate Dictionary is among the best-selling books in American history. Compl. ¶¶ 34, 39. It’s a great read; I recommend it.

Both plaintiffs hold numerous federal trademark registrations for their respective names. Compl. ¶¶ 38, 40.

OpenAI, valued at $730 billion after recent investment rounds, operates ChatGPT, the ubiquitous product with over 900 million weekly users and 50 million consumer subscribers. Compl. ¶ 41.

The Copyright Claims: Three Separate Theories

The complaint structures its copyright infringement theory around three distinct acts of alleged copying, each tied to a different stage of the generative AI pipeline:

First, Britannica alleges that OpenAI copied its copyrighted articles at “massive scale” to train the LLMs that power ChatGPT, using training datasets such as Common Crawl and WebText that crawl and scrape Britannica’s websites. Compl. ¶¶ 50–54. This is the training-stage infringement theory common to virtually every AI copyright complaint filed since 2023.

Second, the complaint identifies a separate act of infringement at the retrieval-augmented generation (“RAG”) stage. RAG is a technique in which an LLM accesses external sources, including live web content, in real time to supplement its pre-trained knowledge base. Compl. ¶¶ 45–47. The complaint alleges that OpenAI copies and uses Britannica’s content through its RAG systems each time ChatGPT retrieves that content to answer a user query. Id. This is analytically distinct from the training-stage claim: it targets real-time copying that occurs every time a user asks ChatGPT a question that triggers a RAG lookup, not just the one-time ingestion of training data.

Third, the complaint alleges that ChatGPT’s outputs themselves infringe Britannica’s copyrights because they contain “full or partial verbatim reproductions” of Britannica’s articles, or text that “resembles, paraphrases, or summarizes” its copyrighted works, or that “mimics the selection and curation of content and lists” in Britannica articles. Compl. ¶ 9.

To support the output-stage claim, the complaint includes side-by-side comparisons. In one example, when prompted to reproduce the body of a Britannica article titled “Education,” GPT-4 generated a near-verbatim copy of the article, omitting some paragraphs but reproducing others word for word. Compl. ¶¶ 58–59. In another, ChatGPT reproduced the identical selection and ordering of quotes from a Britannica article about the Hamilton-Burr duel, including the exact snippets curated by Britannica’s editors, and even noted that Britannica had “fact-check[ed]” the article. Compl. ¶ 63.

The Trademark Claims: Hallucinations as False Designation of Origin

Count V asserts claims under Section 43(a) of the Lanham Act, 15 U.S.C. § 1125, for false designation of origin and trademark dilution. The complaint identifies two distinct forms of trademark harm:

First, false attribution through hallucination. The complaint alleges that ChatGPT sometimes generates fabricated content and “wrongly attributes that text to Plaintiffs using Plaintiffs’ trademarks.” Compl. ¶ 68. In other words, when ChatGPT invents information and presents it as having come from Britannica or Merriam-Webster, it deceives users into believing the hallucinated content carries the authority and accuracy of those brands. Because Britannica’s “reputation rests on accuracy built over more than 250 years,” the complaint argues, associating that brand with AI-fabricated information causes direct reputational harm. Compl. ¶¶ 68, 109.

The second is misleading omission. The complaint alleges that ChatGPT reproduces portions of Britannica’s articles, sometimes verbatim, while silently omitting other portions, “thus falsely giving the impression to users that what ChatGPT presents constitutes the complete reproduction of a Britannica article when, in fact, the content may have notable omissions or changes.” Compl. ¶ 69. The complaint treats these undisclosed omissions as a separate trademark violation because users are misled into believing they are receiving the full, authoritative Britannica text.

Both theories are framed as violations of Section 43(a)(1)(A) (false designation of origin causing likelihood of confusion) and Section 43(c) (dilution by blurring and tarnishment). Compl. ¶¶ 110–111.

The Lanham Act claims address a harm that copyright law does not reach: reputational injury from AI-generated misinformation attributed to a trusted source. A copyright claim can address the unauthorized copying of Britannica’s articles, but it cannot address the situation in which ChatGPT invents information and presents it under Britannica’s name. If the hallucination-as-false-attribution theory survives a motion to dismiss, it would signal that AI developers may face Lanham Act exposure when their models fabricate information and attribute it to a named brand.