Are AI models infringing creators’ copyrights?
Your content — trained on without permission, without payment, without even being told. That’s not hypothetical anymore; it’s being fought out in courtrooms across the US and EU right now, with consequences for every business using AI tools.
Imagine you’ve created the crown jewel of your marketing — a blog that took weeks of research, original images, writing that reflects years of expertise. Now imagine discovering an AI model was trained on it: without your permission, without paying you a rupee, without even telling you. Between 2023 and 2024, over 50 copyright lawsuits were filed against AI companies. By early 2026, the era of “train first, ask later” is officially over — courts, regulators, and creators have pushed back hard, and the legal landscape has changed fundamentally.
copyright lawsuits filed against AI companies in 2023–24
Bartz v. Anthropic settlement — the industry’s largest
Universal Music’s lawsuit against Anthropic over lyrics
Learning from patterns — until it isn’t
When a model trains, it doesn’t read your blog the way a person does. It processes enormous volumes of text, images, code, or audio — absorbing patterns, word relationships, stylistic choices and structure across hundreds of billions of data points, much like a medical student learns principles from case studies rather than memorising each one.
Hire a chef who’s eaten at a thousand restaurants; a dish inspired by those meals isn’t infringing the menus. They’re drawing on learned technique to create something new.
But sometimes the output looks less like inspiration and more like a copied recipe — and that’s where it gets genuinely contentious.
The New York Times showed ChatGPT could reproduce its articles nearly verbatim when prompted. That’s not learning from experience — that’s reproduction.
And it’s exactly why courts worldwide are now drawing new lines around what AI training can and cannot do legally.
The landmark rulings reshaping the law
The most significant copyright settlement in AI history. Authors alleged Anthropic illegally copied their books to train Claude. The court’s nuanced ruling: training on copyrighted books can be fair use — but storing pirated copies to do that training is not. Anthropic’s use of a “shadow library” — downloading books from an unlicensed archive — is where it crossed the line. Settled September 2025 for $1.5 billion, roughly $3,000 per work in the training data.
The Times presented evidence that ChatGPT could reproduce its articles nearly verbatim — so this is about infringing output, not just training-data acquisition. The argument: when a model can reproduce substantial portions of a copyrighted work on request, the “transformative” fair-use defence doesn’t apply. Legal experts call it the most consequential case in the landscape; a ruling against reproduction-as-output would fundamentally change how models are trained and how outputs are filtered. As of mid-2026 it’s still active — watched by every major tech company on the planet.
Universal Music Publishing, Concord, and ABKCO filed a $3.1 billion suit against Anthropic in January 2026, alleging Claude was trained on torrented music — including lyrics from Bruno Mars, the Rolling Stones, and others. BMG Rights Management filed a separate suit. This extends the battle beyond text into music, and the scale signals how seriously the industry treats the threat AI-generated music poses to licensing revenue.
Disney sued Midjourney for “massive” infringement, alleging it reproduces Disney characters. Simultaneously, Disney signed a deal letting OpenAI’s Sora generate videos featuring 200+ Disney characters under a three-year licence — investing $1 billion in OpenAI as part of it. The same properties, in litigation with one platform and a licensing deal with another.
The US Supreme Court declined to hear this in March 2026, effectively confirming the lower-court ruling: AI-generated content cannot receive copyright protection under US law — material must have human authorship. The practical upshot: content created entirely by AI, without meaningful human creative input, is in the public domain. It can’t be owned or protected. Publish AI output as-is, and you hold no copyright over it.
No longer a blank cheque
For years, AI companies assumed training on publicly available internet content was fair use. That’s now being systematically challenged. The US Copyright Office’s May 2025 guidance held that when AI training competes with existing licensing opportunities for the original works, the fair-use analysis tilts against the AI company — if a chatbot can summarise a news article on demand, it competes directly with the publisher’s subscription model. Courts have identified three factors that undermine the defence:
Verbatim reproduction
Once plaintiffs show a model reproduces substantial portions of a work (NYT, Bartz), the “transformative use” argument weakens significantly.
Commercial revenue
Fair use is harder to sustain when the use generates billions in revenue. These aren’t research projects — they’re commercial products competing in the market.
Market harm
When AI output substitutes for the original — summarising articles, mimicking an artist’s style — it causes quantifiable economic harm, tipping the analysis against the AI company.
The takeaway: fair use is not a blank cheque for commercial AI training. Companies that assumed otherwise are now paying — literally.
The EU AI Act
While US courts fight case-by-case, the EU took a structured legislative route. The AI Act entered into force in August 2024 and is largely applicable by August 2026. For general-purpose systems like ChatGPT and Claude, it requires:
- Transparency about training data — companies must document and disclose what training data they used, including copyright-protected works.
- Copyright-compliance summaries — published summaries of content used in training, so rights holders can verify whether their work was included.
- Opt-out mechanisms — rights holders can opt out of having their content used for training, and AI companies must honour it.
For any business operating in Europe or serving European customers, these requirements affect which AI tools you can use and how they must be configured. The GDPR precedent — serious fines for non-compliance — suggests the AI Act will be enforced seriously.
The growing licensing economy
Alongside the litigation, a quieter shift is reshaping how AI and content coexist: licensing. Major publishers increasingly reach agreements with AI companies rather than fighting exclusively in court. OpenAI has deals with Condé Nast, the Guardian Media Group, the Financial Times, and Associated Press; Meta announced a News Corp partnership for Meta AI in March 2026.
The logic is sound — litigation takes years with uncertain outcomes, while a licence generates revenue for the rights holder today, ensures transparency, and gives AI companies legal clarity. The same IP can be an infringement claim against one company and a revenue stream with another.
For businesses holding significant IP, the strategy is active: know what you own, monitor how AI tools use it, and engage in licensing conversations rather than waiting to end up in a lawsuit.
Five practical moves for 2026
Understand how your AI tools were trained
Not all tools carry the same risk. Ask your vendor: what training data was used, and does the company have licensing, consent, or fair-use documentation for it? Enterprise-grade tools often provide training-data transparency; others don’t. It’s a legitimate due-diligence question before signing up.
Remember: AI-generated content has no copyright protection
Following the Thaler confirmation, pure AI output can’t be copyrighted. A competitor can freely copy your AI-generated content — you have no recourse. And AI output alone isn’t a substitute for genuine IP: your original research, team expertise, and brand voice remain protectable.
Human creative input is your copyright protection
The path to protection is human authorship — prompting, editing, selecting, refining, and combining AI with original research and genuine creative judgment. The human contribution is the copyrightable element. It’s also, incidentally, how AI-assisted content performs better.
Watch the outputs, not just the inputs
The NYT case elevated output infringement as a genuine risk. If your tools can reproduce substantial portions of copyrighted text when prompted, you’re potentially exposed. Safeguard: don’t use AI outputs that closely mirror identifiable copyrighted sources — resemblance creates legal risk even if generated independently.
Track India’s response
The framework being set globally will shape Indian policy and enforcement in the coming years — so for businesses generating significant original content, it’s worth watching closely.
No specific AI copyright law yet — but the principles already apply
India doesn’t yet have dedicated AI copyright legislation, though the Copyright Act, 1957 protects original literary, artistic, and musical works — and the principle that copyright exists in the expression of an idea, not the idea itself, applies. Training on copyrighted Indian works without authorisation would raise the same concerns now being litigated in the US and EU.
The IT Ministry has been developing an AI governance framework, and the Copyright Office is closely monitoring international developments. Indian creators, journalists, authors, musicians, and publishers are watching these cases with real interest — because the global framework will shape India’s rules and enforcement in the years ahead.
AI and copyright can co-exist — but only on fair terms.
AI isn’t going to collapse under copyright pressure — the tools are genuinely transformative and will stay central to how businesses operate. But the terms on which AI and copyright coexist are being rewritten right now, in courtrooms, regulators, and licensing negotiations. The era of “train on everything, sort out the law later” is over; in its place: licensing where possible, transparency about training data, respect for opt-outs, and real liability for infringing outputs.
The right question isn’t “should I use AI?” — of course you should. It’s “am I using it responsibly, with an honest understanding of where the legal lines are?”
Know your tools and their training-data posture. Protect your original content and keep human authorship clear in what you publish. Watch the cases moving through US and EU courts — their outcomes will define the environment you operate in. AI and human creativity work best together; the rules for how are being written right now, so make sure you’re on the right side of that conversation.
Integrate AI ethically — and stay on the right side of the law.
Ready to bring AI into your business responsibly? Let’s talk about using AI tools smartly, in a way that protects your brand and your IP.
This article is general information, not legal advice. AI copyright law is evolving rapidly across jurisdictions — confirm specifics with qualified counsel before acting.
