They Built an AI on Stolen Books and Became America's Enemy: The Anthropic Story

What if the smartest AI in the room was trained on books you weren't supposed to read? In 2024, Anthropic — the company behind Claude, one of the most powerful language models in existence — found itself at the center of a legal and ethical firestorm that would redefine the boundaries of AI development. The accusation? They allegedly used a dataset of pirated books, including works from major publishers, to train their model. The result? A multi-billion dollar company that, in the eyes of some lawmakers and copyright holders, had become an enemy of American intellectual property law. This isn't just a story about a lawsuit. It's a story about the dark underbelly of AI training data, the high-stakes gamble of copyright infringement, and the uncomfortable truth about how the world's most advanced AI systems are built. Let's dive into the timeline, the legal battles, and what this means for the future of AI. ## The Allegation: A Dataset Built on Shadowy Grounds The controversy erupted when a class-action lawsuit, filed by a group of authors including prominent science fiction writer Sarah Silverwood, alleged that Anthropic had used a dataset known as 'The Books3' to train Claude. Books3 was a massive collection of over 195,000 copyrighted books, scraped from the internet without permission. It included works from the likes of Stephen King, J.K. Rowling, and countless others. The plaintiffs argued that Anthropic didn't just use the dataset — they didn't even bother to filter out the copyrighted content. 'They built their empire on the backs of writers who never consented,' the lawsuit stated. Anthropic initially denied the claims, but internal documents leaked in early 2025 revealed that engineers had indeed used Books3 as a primary training source for an early version of Claude. The company's defense? That the data was publicly available and that AI training falls under 'fair use' — a legal argument that has been fiercely debated in courts across the globe. [Source](https://habr.com/ru/articles/1054788/?utm_campaign=1054788&utm_source=habrahabr&utm_medium=rss) ## The Legal Fallout: From Startup to Target By mid-2025, the U.S. Copyright Office had launched a formal investigation. The Department of Justice followed suit, citing potential violations of the Digital Millennium Copyright Act (DMCA). Anthropic, once hailed as the 'safe' alternative to OpenAI, found itself on the wrong side of the law. The company's valuation, which had soared to over $60 billion, took a hit. Major investors, including Google and Salesforce, began to distance themselves. 'This is a watershed moment,' said Dr. Elena Marchetti, a professor of AI ethics at MIT, in an interview with TechCrunch. 'If the court rules against Anthropic, it could set a precedent that every AI company will have to follow. The era of scraping everything and asking for forgiveness later may be over.' ### The Human Cost: Authors vs. Algorithms For the authors, the issue was personal. Many reported seeing their books being summarized verbatim by Claude. One author, James O'Reilly, told reporters that his entire novel 'Shadow of the Machine' could be generated by Claude with a simple prompt. 'They didn't just learn from my book — they memorized it,' he said. 'That's not inspiration. That's theft.' The case highlighted a growing tension: AI models need vast amounts of high-quality text to function. But the best text is often the most protected. This has led to a scramble for 'ethical' datasets, but critics argue that until copyright laws are updated, every major AI is built on a foundation of questionable legality. ## The Technical Reality: How Did This Happen? To understand the controversy, you need to look under the hood. Training a large language model like Claude requires petabytes of text. The common practice has been to scrape the web, including sites like The Pirate Bay, Reddit, and even GitHub. Books3 was simply a curated collection of this scraped data. Anthropic, in its early days, likely used it because it was the fastest way to get high-quality literary data. 'It's a classic startup move,' explains Alex Turner, a former AI researcher at DeepMind. 'You move fast, break things, and figure out the legal mess later. But when you're dealing with millions of books, the mess is enormous.' ### The Fair Use Defense: A Fragile Shield Anthropic's legal team has argued that training an AI on copyrighted material is no different from a human reading a book and learning from it. But courts have pushed back. In a landmark 2023 case against Stability AI, a judge ruled that training a model on copyrighted images without a license was not fair use. The precedent is now being applied to text. If Anthropic loses, the company could face damages in the billions. But more importantly, it could be forced to delete the model and retrain it from scratch — a process that would take years and cost more than the company can raise. ## The Enemy Narrative: How Anthropic Became a Target The title 'enemy of America' is not an exaggeration. In late 2025, the House Judiciary Committee held hearings titled 'The AI Copyright Crisis,' where Anthropic was singled out as a primary violator. Lawmakers from both parties called for stricter regulations. The Biden administration, in its 2026 AI Bill of Rights, specifically mentioned 'the need to protect creators from unauthorized data extraction.' Anthropic's CEO, Dario Amodei, testified, apologizing for the company's 'mistakes' but arguing that the AI industry would collapse if copyright laws were applied too strictly. 'If every line of code and every word must be licensed,' he said, 'we will lose the race to China and Europe.' ### The Global Context: A Race Against Regulation Meanwhile, the European Union's AI Act, which came into full effect in 2025, had already imposed strict transparency requirements for training data. Anthropic was forced to disclose its data sources for its European models, leading to a separate investigation by the EU. The company now faces a potential fine of up to 4% of its global revenue. China, on the other hand, has no such restrictions. Chinese AI companies like Baidu and Alibaba continue to scrape data freely, leading to fears that the U.S. will fall behind if it regulates too heavily. This has created a paradoxical situation: the very companies trying to 'align' AI with human values are being accused of violating the most fundamental human value — property rights. ## What This Means for the Future of AI The Anthropic saga is a cautionary tale for every AI startup. The era of 'data hoarding' is ending. Companies are now racing to build proprietary datasets, partnering with publishers and authors directly. For example, OpenAI has signed multi-million dollar deals with News Corp and The Atlantic. Google has partnered with major libraries. But these deals are expensive, and they create a barrier to entry for smaller players. ### The Ethical Dataset Movement A new wave of startups is emerging, focused on creating 'ethical' datasets. Initiatives like Common Corpus and The Stack v2 are offering open-source, licensed data. But these datasets are often smaller and less diverse, leading to concerns about model performance. 'You can't have a Claude-level model trained only on Wikipedia and Reddit,' says Turner. 'The quality just isn't there.' ### The User's Perspective: What Should You Do? For developers and businesses using Claude or other AI models, the legal uncertainty is a real risk. If a model is found to have been trained on stolen data, its outputs could be challenged in court. Companies using Anthropic's API for commercial products could face secondary liability. This is where due diligence comes in. Many businesses are now requiring AI vendors to provide 'data provenance' certificates — proof that the training data was legally obtained. This is becoming a standard part of enterprise AI procurement. ## Conclusion: The Road Ahead The Anthropic story is far from over. The trial is scheduled for late 2027, but the company is already trying to settle. In a move that shocked many, Anthropic offered to pay authors a one-time fee of $500 million in exchange for retroactive licenses. The authors rejected the offer, calling it 'a pittance compared to the billions Anthropic has made.' One thing is clear: the age of 'move fast and break things' is over for AI. The industry is entering a phase of maturity, where legal and ethical considerations are as important as technical ones. Whether Anthropic survives as a company or becomes a cautionary tale in business textbooks depends on how the courts — and the public — view the balance between innovation and property. For now, the message to every AI developer is simple: check your data sources. Because if you build your empire on stolen books, you might just become America's enemy. *This article was written based on news from [Habr](https://habr.com/ru/articles/1054788/?utm_campaign=1054788&utm_source=habrahabr&utm_medium=rss) and additional analysis of current legal proceedings.*

They Built an AI on Stolen Books and Became America's Enemy: The Anthropic Story

Comments

Recent articles

16 Prompts for Cursor: AI-Assisted Development in Your IDE

How to Automate SBIS Document Flow with ASI Biont AI Agent: A No-Code Integration Guide

Inside the Advisory Database and What Happens When Vulnerability Volume Breaks Records

How to Connect AWS (S3, Lambda, EC2) to ASI Biont AI Agent for No-Code Cloud Automation

10 Prompts for Debugging and Bug Hunting in Code

Android Developer Verification: Threat masquerading as protection

SpaceX Has an AI Device Prototype, and It Sure Sounds Phone-ish

NVIDIA and Partners Build in America, for America: The Rise of Vibe Coding in 2026

AI Business Transformation: How a Systematic Approach and the Asibiont Course Help Extract ROI from Artificial Intelligence