Publishers push back on AI

Publishers and authors are tightening access to their work as the fight over artificial intelligence training data shifts from protest to payment demands. News companies have started blocking or limiting crawlers, and agents and editors say they are also spending more time sorting possible machine-written submissions. (niemanlab.org) The immediate flashpoint is web scraping: software bots copy pages at scale so artificial intelligence systems can learn patterns from books, articles, and images. OpenAI says website owners can use robots.txt controls for GPTBot and OAI-SearchBot, but publishers increasingly want licenses, not just opt-out tools. (openai.com) That stance has hardened as publishers argue the traffic tradeoff is collapsing. A March 2025 TollBit report cited by News Media Alliance said click-through rates from artificial intelligence chatbots were 95.7 percent lower than traditional Google search, with a referral rate of 0.37 percent. (newsmediaalliance.org) Some publishers are now restricting even the Internet Archive, whose Wayback Machine has long preserved public webpages. Nieman Lab reported in January 2026 that publishers including The Guardian limited Internet Archive crawling after concluding archived copies could become a back door for artificial intelligence developers seeking training data. (niemanlab.org) The publishing industry’s legal posture has also sharpened. On February 10, 2026, a group of independent publishers coordinated by the Independent Publishers Guild sent letters of claim to Anthropic, Meta, OpenAI, Google, and xAI seeking details on datasets, legal bases for use, and undertakings to stop unauthorized use. (foxwilliams.com) Trade groups have spent the past year building a broader coalition around that message. The Association of American Publishers said in October 2024 that it joined more than 10,000 creators and partner organizations in condemning unlicensed use of creative works for generative artificial intelligence training. (publishers.org) Court cases have given publishers and authors fresh leverage. The Association of American Publishers said a proposed settlement in Bartz v. Anthropic requires Anthropic to pay $1.5 billion to authors and publishers over books downloaded from pirate sites, with preliminary approval announced in September 2025. (publishers.org) At the same time, editors say the problem is arriving through the front door. CBC reported in March 2024 that literary agents and publishers were seeing a rise in suspected artificial intelligence-generated manuscripts after the viral exposure of “The Last of the Moon Girls” author’s pseudonymous novel “Shy Girl,” complicating screening, originality checks, and contract questions. (cbc.ca) Those rights questions are practical as much as philosophical. The United States Copyright Office has repeatedly said copyright protects human authorship, not purely machine-generated material, which means publishers acquiring manuscripts now have to ask not only whether text is good enough to buy, but whether the author can fully grant the rights being sold. (copyright.gov) Artificial intelligence companies have argued in court that training on copyrighted works can be fair use, and some media companies have signed licensing deals rather than sue. But the direction inside publishing is moving toward a simpler demand: if developers want books and journalism to build products, they should pay for them before they take them. (cjr.org)