Canadian publishers are in a tough spot. How do you protect your copyrighted work when large language models (LLMs) like Anthropic’s Claude are trained by scraping up everything on the internet, usually without asking for a license or paying a dime? The collision of new tech and old IP rights is kicking off major AI legal fights, forcing everyone to rethink what copyright even means anymore. In the age of generative AI, it’s an open question if publishers can actually protect their work.
Key Takeaways
- Publishers are finally suing AI developers, saying their content was used for training without permission, which is copyright infringement.
- The whole fight boils down to “fair dealing” (in Canada) or “fair use” (in the US). AI companies say their training is far-reaching. Publishers say it’s killing their business.
- New ways to get paid are popping up, like specific licensing deals for AI and tech like content fingerprinting or watermarking to track where content goes.
- Whatever the courts decide in these first big lawsuits will set the rules for how AI models can use copyrighted material and what future licensing deals will look like.
- Publishers need to get proactive: check who’s scraping your site, look into digital rights management, and think about banding together to negotiate with AI firms.
The Unseen Data Ingestion: A Publisher’s Dilemma
The root of the problem for Canadian publishers is pretty straightforward: their expensive-to-produce articles, books, and research are being hoovered up by AI models without a license or a check. To learn language and facts, LLMs need to read a staggering amount of text, so developers just scrape the public web, assuming anything they can see is free to take. It’s great for building AI quickly, but it completely torpedoes the business model for anyone who creates content for a living. Think about a small independent publisher in Vancouver that spends months on an investigative report about local corruption. When an LLM eats that content and spits it back out as a summary, the publisher who paid for the original work gets nothing, just a cheaper, AI-generated competitor diluting their value.
The Canadian Copyright Act, with its rules in sections 27 and 28, was written decades before anyone dreamed of generative AI, leaving a huge legal grey area. Publishers are arguing that the simple act of copying their work to train a model is infringement, period. It doesn’t matter if the final AI output is a word-for-word copy. The initial unauthorized copy deprives them of control and money. The money side of this is brutal. If AI can spit out articles that are good enough to compete with human writers, why would anyone pay for licensed content? This could wreck ad revenue, subscriptions, and direct sales. This is already happening. News outlets all over the world are seeing their referral traffic from search engines drop as AI summaries get baked into the results pages.
What Went Wrong First: The Inaction Trap
At first, a lot of publishers just sat back and watched, totally underestimating how fast and big this AI thing would get. The thinking was that AI would be a handy tool for simple tasks, not something that would compete with their core product or threaten their IP. So they did nothing, and their valuable digital archives were left wide open for AI scrapers without any real technical or legal fences. Some even made it easier for bots to grab text with generic site structures. Nobody wanted to get into a money-draining legal fight with a tech giant, which just gave the AI developers a huge head start.
The other big mistake was not banding together to negotiate as a group or set some industry ground rules from the start. A single publisher trying to make a deal with a huge AI firm has almost no use. Without a united front, AI companies could just set their own terms or, more often, just take the data without asking, knowing there was a ton of it out there for the scraping. The focus was stuck on catching direct plagiarism, completely missing the bigger problem of their entire archives being used for training. Being reactive instead of proactive let the AI industry grow up with a business model based on getting its raw materials for free.
The Legal Offensive: Publishers Take a Stand
The wake-up call came when publishers realized this was a fight for their survival. In 2025, a group of Canadian news publishers, including big names like The Globe and Mail and Torstar, sued Anthropic in the Federal Court of Canada for copyright infringement. They are specifically calling out the company for copying and adapting their articles to train the Claude models. This lawsuit is a lot like the one The New York Times Company filed against OpenAI and Microsoft in late 2023, which laid out in detail how its reporting was used to build ChatGPT.
The publishers’ lawsuits are built on a couple of key points. First, they’re saying that just copying the content into a training set is illegal reproduction under copyright law, no matter what the AI spits out later. Second, they claim these AI models act as a direct substitute for their journalism by summarizing or rewriting their work, which devalues the original product. This argument gets really strong when an AI answers a question that would have previously sent a user directly to a publisher’s site for the answer.
Anthropic and the other AI developers will almost certainly use a “fair dealing” defense (that’s Canada’s version of “fair use”). Their argument will be that training an AI is far-reaching because the model learns patterns, it doesn’t just memorize text. They’ll also probably say it’s for the public good, making information more accessible, and that blocking their access to data will kill innovation. But publishers have a solid counter: these are commercial products, not academic projects, and they compete directly with us which blows a hole in any “fair dealing” claim.
These fights are messy. They require expert witnesses to explain how LLMs actually work, what “copying” even means in a digital world, and the real economic damage to the publishing industry. A smoking gun for publishers is showing examples where an AI spits out big chunks of their copyrighted text nearly verbatim. Or even better (for their case), when a Claude model reproduces factual errors that could only have come from their original reporting, proving it was copied directly from their work.
Beyond the Courts: Licensing and Technological Solutions
While the lawyers are busy, publishers are also trying to find their own solutions. The biggest move is a push for new licensing agreements built just for AI companies. The idea is to create a system where AI developers have to pay to license content for training, just like they would for stock photos or music. This would give publishers a new way to make money and some control over how their work is used. Groups like the News Media Alliance are already trying to build industry-wide frameworks for this.
On the tech side, publishers are starting to use smarter digital rights management (DRM) tools. This can be as simple as changing their website’s robots.txt file to block crawlers from known AI companies. Others are trying out content fingerprinting and watermarking, which embed invisible codes in their work. These markers could, in theory, let a publisher see when their content has been scraped by an AI or even spot traces of their work in an AI’s output, which would be powerful evidence in a lawsuit.
Some publishers are also trying partnerships. Instead of just saying no, they’re cutting deals to license their content to AI developers with strict terms on attribution, quality, and of course, payment. This is a pragmatic approach that accepts AI is here to stay but tries to build a future where people can still get paid for creating original work. For example, you could see a Canadian academic publisher licensing its entire journal archive to an AI firm that’s building a specialized medical LLM, with terms that guarantee proper citation and a share of the revenue.
All this fighting and tech development is creating a totally new environment. Publishers are getting more aggressive about protecting their IP, and AI companies are (slowly) realizing they might have to actually pay for their training data. It’s a messy negotiation, but one that has to happen for either of these industries to have a future. My take is that AI companies, especially the ones with billions in funding, need to start treating content licensing as a basic cost of doing business, just like paying for servers or electricity. It’s not some barrier to progress. In the long run, good AI needs good original content, and that requires a healthy industry of people who create it.
The Future Field: Precedents and Partnerships
The first rulings from these AI legal battles will set major precedents. If the courts side with publishers, AI developers will have to completely rethink how they get their data. They’d have to start investing serious money in licensing deals which would create a whole new market for ethically sourced training data. If the courts side with the AI companies, it could gut copyright protection for the digital age and make it incredibly hard for anyone creating content to make a living.
Outside the courtroom, things are already shifting toward collaboration. A few AI companies have started announcing partnerships with publishers to license content. These deals aren’t just about a one-time payment. They often include data-sharing agreements where publishers get to see how their content is being used, or they get access to AI tools to improve their own work. It’s a sign that the relationship is moving from pure conflict to something more complicated.
For Canadian publishers, how this all shakes out will determine if they can keep investing in quality journalism. It’s that simple. If they can defend their rights, it proves that human-made content still has value, even when AI is everywhere. If they can’t, the industry is in for a world of pain. The goal has to be finding a balance where AI innovation can happen without burning down the house that intellectual property built.
Nothing is settled yet, but the fact that publishers in Canada and elsewhere are fighting back is forcing the issue. The future is probably going to be a mix of things: ongoing court cases to draw clear lines, combined with new licensing deals and better tech to protect content. Publishers have to stay on their toes and be ready to change tactics as the tech and the law keep changing.
The fights between Canadian publishers and AI firms like Anthropic are making one thing clear: we need answers on how copyright law applies to AI training. To survive, publishers need to be in the legal fight while also building new licensing models that ensure original content has a future. It’s time to audit your site for AI scraping and look into what a collective licensing deal could do for you.
What’s the main legal argument from Canadian publishers against AI companies?
They’re arguing that AI companies are breaking copyright law. They do this by copying articles and other work without permission to train their language models, which makes the original work less valuable and creates a direct competitor.
How do AI companies defend themselves against these copyright claims?
Their main defense is “fair dealing” (in Canada) or “fair use” (in the US). They claim their use of the material is for a greater purpose, training a model to learn, and that it helps the public without directly replacing the original work.
What tech are publishers using to protect their content from AI?
They’re using things like robots.txt files to block AI web crawlers, trying out content fingerprinting and watermarking to track where their stuff ends up, and generally beefing up their digital rights management to lock down their data.
What are the possible outcomes of these lawsuits for the publishing industry?
It could go two ways. A win for publishers could force AI companies to pay for training data, creating a new revenue source. A loss could weaken copyright law and cause huge problems for traditional publishing business models.
Can publishers just license their content to AI companies?
Yes, and that’s becoming a popular solution. Publishers are starting to negotiate deals with AI developers to license their content for training. These deals set clear terms for payment, attribution, and quality control.