Merriam-Webster and Britannica Sue OpenAI Over AI Training Data Theft

The legal landscape for generative artificial intelligence has reached a critical inflection point as two of the world’s most venerable knowledge institutions, Merriam-Webster and Britannica, have initiated litigation against OpenAI. The lawsuit, filed in March 2026, alleges that OpenAI’s ChatGPT was trained on vast quantities of proprietary material stolen from their digital archives. This case represents a significant escalation in the conflict between legacy intellectual property holders and the rapid expansion of large language models (LLMs), moving beyond creative prose and news into the foundational realm of factual definitions and reference data.

At the heart of the complaint is the concept of digital cannibalization. Merriam-Webster and Britannica argue that by ingesting their curated definitions and historical entries, ChatGPT has effectively become a substitute for their own platforms. When a user asks an AI for a definition or a historical summary, they no longer need to visit the source websites, which rely heavily on ad revenue and subscriptions. The plaintiffs contend that OpenAI is essentially 'laundering' their high-quality, human-vetted data through a machine-learning process to create a competing product that directly undermines the original creators' business models. This argument mirrors the 'substitution' theory seen in previous copyright cases, where a new technology is found to be infringing because it serves as a direct market replacement for the original work.

“

The legal landscape for generative artificial intelligence has reached a critical inflection point as two of the world’s most venerable knowledge institutions, Merriam-Webster and Britannica, have initiated litigation against OpenAI.

Industry context suggests this lawsuit is part of a broader 'IP rebellion' against the Silicon Valley ethos of 'move fast and break things.' For years, AI developers have relied on the 'fair use' doctrine, arguing that training a model is a transformative process that creates something entirely new. However, the Merriam-Webster suit challenges this by highlighting that the output of ChatGPT often mirrors the specific phrasing and structural nuances of their dictionary entries. Unlike the New York Times lawsuit, which focused on the reproduction of journalistic articles, this case focuses on the very building blocks of language and factual synthesis, raising complex questions about whether a definition—which is inherently factual—can be protected under copyright law when it is the result of specific editorial curation.