AI Data Broker for Publishing: License Books for AI Training

If you run a publishing house, a magazine, or a news outlet, you have probably asked some version of this: AI companies are scraping our content, so can we get paid for our archive instead? Decades of backlist titles, back issues, and articles are sitting in files and databases, earning little or nothing.

Meanwhile the lawsuits keep coming. Canadian news publishers are suing OpenAI in Ontario, and class actions over AI training have been filed in Canada against Anthropic and Meta. In the United States, a court approved a US$1.5 billion settlement in a case alleging Anthropic trained on illegally obtained copies of books.

Litigation is one path, and it is for your lawyer to advise on. Licensing is the other. Clean, complete, well-formatted text with clear rights is what licensed buyers pay for, because it is safer and more useful than anything they can scrape.

The buyers are the companies building large language models, and the deals so far set a useful range. HarperCollins offered authors a three-year AI training licence with Microsoft at US$5,000 per book, split evenly with the author, and authors could accept or decline. Wiley said it expected to earn US$44 million from AI rights deals. News Corp's agreement with OpenAI could be worth close to US$250 million over five years, according to The Wall Street Journal.

Litigation sets another benchmark. The Anthropic settlement approved in July 2026 works out to about US$3,000 per book across more than 400,000 works. That is a legal remedy, not a market price, but it shows how seriously courts are taking the difference between licensed and unlicensed copies.

Smaller publishers will not see those totals. Value depends on volume, uniqueness, format, and how cleanly the rights can be shown. A specialist backlist with clear contracts can be more attractive than a larger list with gaps in its paperwork.

Canadian publishers are already pushing back on unlicensed training. A coalition including The Canadian Press, Torstar, The Globe and Mail, Postmedia, and CBC/Radio-Canada is suing OpenAI, and in November 2025 the Ontario Superior Court ruled the case can proceed in Ontario. That was a ruling on jurisdiction, not the merits.

Canada's Copyright Act has no text and data mining exception. In a May 2026 submission, Access Copyright and Copibec asked the federal government not to add one, and to require AI developers to be transparent about training data. The federal AI strategy released in June 2026 did not address copyright, so for now the practical route to getting paid is a direct licence.

Canada's bilingual publishing base is an advantage. French-language and paired French and English text is scarcer than English and useful to developers building models for Canadian and francophone users. For publishers, that is a form of data monetization most other markets cannot offer.