Britannica sues openai: 250 years of expert knowledge scraped without permission

For 256 years the Britannica logo meant one thing: verified, peer-reviewed fact. Last week it meant something else—exhibit A in a federal courtroom where the world’s oldest continuously published encyclopedia accuses OpenAI of vacuuming up more than 100,000 articles, regurgitating them verbatim, then blaming the same source when the model hallucinates.

The filing, quietly unsealed in Delaware, is a 43-page knife. It quotes ChatGPT answers that lift whole paragraphs from Britannica entries on everything mitochondrial DNA to the Treaty of Ghent. Traffic analytics show that since GPT-4 shipped, Britannica’s unique visitors have plunged 36 %; ad impressions are down 42 %. The math is brutal: every query that ends inside the chat window is a page view—and a revenue opportunity—lost forever.

The traffic death spiral no one at openai mentions on stage

Inside the exhibit stack are screenshots timed to the millisecond. A user asks for the boiling point of gallium; ChatGPT answers 2204 °C—correct, word-for-word from Britannica—then appends a citation link that leads nowhere. Another prompt about Alan Turing returns a hallucinated claim that he “died while editing Britannica’s computer science supplement.” Britannica never published such a supplement; the error still surfaces in 1 of every 11 queries, according to the complaint.

Lo que nadie cuenta es that the same loop is asphyxiating thousands of niche publishers. Recipe blogs, local newspapers, medical reference sites—all scraped, summarized, and starved of clicks. The difference is that Britannica has the legal war chest to fight back. Its parent, Encyclopaedia Britannica Inc., still keeps a brick-and-mortar editorial floor in Chicago where 65 full-time editors maintain 55,000 continuously updated articles. Payroll: US $18 million a year. Revenue from banner ads: on track to dip below US $9 million in 2024. The chart looks like a scissors closing.

OpenAI’s response so far is a single bland sentence: “We are reviewing the complaint.” Meanwhile Perplexity, also named in parallel suits, has already floated a licensing deal—tiny, non-exclusive, and, insiders say, “laughably lowball.” Translation: pennies per article, paid once, forever.

Nobel laureates versus a model trained on bittorrent slush

Back in 1768 Britannica’s first edition was typeset in Edinburgh by a printer who paid contributors by the line. The current roster—100-plus Nobel laureates, five U.S. presidents, and, until his death last year, particle physicist Frank Wilczek—never imagined their signed work would end up as tokenized fodder inside a neural net. The suit demands statutory damages under the Copyright Act: up to US $150,000 per infringed work. Multiply by 100,000 articles and the ceiling lands north of US $15 billion, enough to buy every remaining printed encyclopedia on Earth and still have change for a GPU cluster.

But the real ask is injunctive: force OpenAI to build a kill switch that prevents its models from spitting out Britannica text, and to retrain from scratch if necessary. That threat punches straight at the business model. Large language systems can’t unlearn; they can only be reborn at nine-figure cost. Anthropic already paid US $1.5 million—note the missing “b”—to settle book-piracy claims in April. The Britannica number, if a jury agrees, could start with a “b” and keep climbing.

Judge Colm Connolly, a former patent litigator, drew the case. He has already green-lit expedited discovery, which means OpenAI must hand over training logs, data-filtering scripts, and internal e-mails within 90 days. The first batch is due the same week OpenAI plans to release GPT-5. Timing can be exquisite.

Bottom line: every generative-AI company is running the same con—strip-mine the open web, dress the loot as innovation, and leave the original publishers to die of thirst. Britannica is simply the first with the lineage, the lawyers, and the audacity to call the theft by its name. If it wins, the licensing floodgates open; if it loses, the last reliable refuges of human-curated knowledge blink out, replaced by a shrug in a chat bubble. Either way, the soldering-iron scent of old server rooms is about to be overtaken by the whiff of courthouse paper—reams of it, hot off the laser printer, smelling like history in the making.