Britannica says OpenAI scraped thousands of its online articles to train its AI models, which now produce responses that closely mirror the publisher's content.