AI in the Courtroom: The Legal Cases That Will Define the Next Decade

The AI industry was built in a legal vacuum. That vacuum is closing fast. A dozen major lawsuits are working their way through courts in the US, UK, and EU. Regulatory frameworks that existed only on paper are getting their first real enforcement actions. And the outcomes of these cases won't just affect the companies named in the filings — they'll set the rules for an entire industry. I've spent twelve years covering tech policy, and I've never seen a legal docket this consequential. Here are the cases that matter most. ## The New York Times v. OpenAI: The Big One Filed in December 2023 in the Southern District of New York, The New York Times v. OpenAI and Microsoft is the most important copyright case in AI. Full stop. The Times alleges that OpenAI trained GPT models on millions of copyrighted articles without permission or payment, and that ChatGPT can reproduce Times content near-verbatim when prompted. The complaint includes exhibits showing ChatGPT outputting passages from Times articles word-for-word — a finding that cut through the abstract legal arguments and gave the case visceral impact. OpenAI's defense rests on fair use — the same doctrine that allowed Google to scan millions of books for Google Books Search (upheld in Authors Guild v. Google, 2015). The argument: training on copyrighted material is transformative because the model doesn't store or regurgitate the original works; it learns statistical patterns from them. The outputs are new works, not copies. The Times counters that the near-verbatim reproduction proves the models do, functionally, store copyrighted content. And that the commercial purpose — OpenAI charges for ChatGPT subscriptions — weakens the fair use argument. Courts weigh four factors in fair use analysis: purpose and character of use, nature of the copyrighted work, amount used, and effect on the market. The Times argues it loses on all four. Where it stands in early 2026: the case has survived initial motions to dismiss. Discovery is ongoing, which means OpenAI may have to reveal details about its training data, data sourcing practices, and internal communications about copyright. If the Times gets access to OpenAI's internal emails about training data decisions, the case could get very uncomfortable very quickly. If the Times wins — or even gets a favorable settlement — it'll establish that training on copyrighted content requires licensing. Every AI company that scraped the web for training data would face massive retroactive liability. The economic implications are almost incalculable. If OpenAI wins on fair use, it'll cement the legal foundation for AI training as it currently exists. Every other copyright case against AI companies would be weakened. My prediction: settlement. Both sides have too much to lose from a full trial. The Times doesn't want to risk a precedent that fair use covers AI training. OpenAI doesn't want discovery to expose how aggressively they sourced training data. A licensing deal — probably in the hundreds of millions — is the most likely outcome. But that settlement will effectively set a market price for training data licensing, which has its own massive implications. ## Getty Images v. Stability AI: The Visual Arts Front Filed in February 2023, Getty's case against Stability AI is the NYT case's visual equivalent. Getty alleges that Stability AI used millions of Getty-watermarked images to train Stable Diffusion without licensing. Early versions of Stable Diffusion literally reproduced Getty watermarks in generated images — a smoking gun that's hard to argue around. This case is playing out in both US courts (District of Delaware) and UK courts simultaneously. The UK proceeding is particularly important because UK copyright law is different from US fair use doctrine. The UK has a narrower "fair dealing" exception, which doesn't cover commercial use as liberally. If Getty wins in the UK, it could create a precedent that forces AI companies to license training data for European and UK markets regardless of what US courts decide. Stability AI has struggled financially throughout the litigation — losing key staff, burning through cash, and cycling through leadership. Whether the company survives the lawsuit in its current form is an open question. But the legal precedent will outlast the company. ## The EU AI Act: From Paper to Practice The EU's Artificial Intelligence Act entered into force on August 1, 2024. It's the world's first comprehensive AI regulatory framework, and it's now moving from theory to enforcement. The Act classifies AI systems by risk: unacceptable (banned), high-risk (heavy compliance), limited risk (transparency requirements), and minimal risk (no regulation). There's also a separate category for general-purpose AI models — which is where things get interesting for the foundation model companies. Under the Act, providers of general-purpose AI models must publish sufficiently detailed summaries of their training data. This is a direct response to the opacity that characterizes most AI training processes. OpenAI, Anthropic, and Google have all been vague about exactly what data they use. The EU says that's not acceptable anymore. For models that pose "systemic risk" — essentially, any frontier model trained above a certain compute threshold — the requirements are stricter: model evaluation, adversarial testing, incident reporting, and cybersecurity measures. The enforcement timeline is staggered. Provisions banning unacceptable AI practices (social scoring, manipulative AI) took effect in February 2025. General-purpose AI obligations kick in around August 2025. High-risk system rules apply by August 2026. Here's what makes the EU AI Act different from US regulation (which barely exists at the federal level): it applies extraterritorially. If your AI system has users in the EU, you must comply — even if your company is based in San Francisco. This is the GDPR playbook. And just like GDPR forced global changes in data practices, the AI Act will force global changes in AI development. The first enforcement actions are expected in mid-2026, and the European Artificial Intelligence Board — established by the Act to coordinate national enforcement — is already working. Which company gets the first major fine is anyone's guess, but my money's on a company that fails the training data transparency requirements. That's the easiest violation to identify and the one the EU has been most vocal about. ## Thomson Reuters v. Ross Intelligence: The Precedent Setter This one flew under the radar but matters. Thomson Reuters sued Ross Intelligence (an AI legal research startup) for training on Westlaw content. In February 2025, a Delaware court ruled that Ross's use wasn't fair use, finding that copying headnotes and key numbers from Westlaw to train an AI system wasn't transformative. This is the first major court ruling directly addressing whether training an AI model on copyrighted content constitutes fair use — and the answer was no. The ruling is narrow — it's a district court decision, not binding precedent outside Delaware — but it signals how courts might approach the larger cases. If more courts follow this reasoning, the "training is fair use" argument that OpenAI and others rely on gets much harder to make. ## Andersen v. Stability AI: The Class Action Sarah Andersen, Kelly McKernan, and Karla Ortiz filed a class action against Stability AI, Midjourney, and DeviantArt in January 2023, alleging that AI image generators were trained on billions of copyrighted artworks without consent. The case has been partially dismissed and partially revived multiple times. What's survived is the direct copyright infringement claim — the argument that AI-generated images that closely resemble specific copyrighted works are infringing copies. This case matters less for the specific legal arguments and more for the class it represents. If certified, it would establish that individual artists have standing to sue AI companies as a class, potentially opening the door to billions in damages across the industry. ## What's Coming The legal landscape for AI is going to get much messier before it gets cleaner. Expect: **More copyright cases.** Record labels are already suing AI music generators. Authors are suing over book training data. Photographers, journalists, and software developers all have potential claims. Every creative industry that contributed to AI training data is evaluating litigation. **Privacy enforcement.** The EU's GDPR is increasingly being applied to AI training data. Italy temporarily banned ChatGPT in 2023 over privacy concerns. More GDPR actions targeting AI companies are coming, particularly around the right to be forgotten and consent for training data. **Product liability.** When an AI makes a medical recommendation that harms a patient, or a legal recommendation that costs a client a case, who's liable? The model provider? The app developer? The user? These cases haven't been filed yet, but they will be. And the answers aren't obvious. **Regulatory divergence.** The EU has the AI Act. The US has... executive orders that can be revoked by the next administration (which the Trump administration has already done). The UK is pursuing a "pro-innovation" light-touch approach. This divergence creates compliance nightmares for global companies and potential regulatory arbitrage. ## The Stakes I want to be clear about what's at stake here. These cases aren't just about money — though the financial exposure runs into the hundreds of billions. They're about who owns the inputs and outputs of AI systems. Whether creative workers get compensated when their work trains the models that may replace them. Whether companies can be held accountable for what their models do. Whether "we scraped the internet" is a business strategy or a legal liability. The answers will shape the AI industry for decades. And unlike most tech regulation, which arrives years after the technology has been deployed, this wave of litigation is happening in real time. The models are still being trained. The products are still being built. And the legal ground under them is shifting with every filing. Pay attention to the courtrooms. That's where the rules of AI are actually being written.