Author :- Babli, a student at Lloyd School of Law, Greater Noida
Abstract
Artificial intelligence has burst out of research labs and into courtrooms, contracts, and legal rulebooks far faster than most laws can keep up. Generative AI is now writing legal briefs, drafting patent applications, screening job applicants, and handling customer questions- tasks that keep bumping up against long- standing legal rules built entirely around human beings.
This article takes a close look at how copyright, patent, tort, and regulatory laws are being stretched and reshaped to deal with AI. It draws on recent court decisions from the United States, United Kingdom, Canada, and the European Union, as well as India’s developing regulatory approach. At its heart, the piece argues that judges aren’t waiting around for lawmakers to pass brand- new AI rules. Instead, they’re taking familiar legal principles- like fair use negligent misrepresentation, and the requirement of human authorship- and applying them to these new, unfamiliar situations. The result is a growing patchwork of precedents that businesses, lawyers, and creators simply can’t afford to ignore.
To the Point
Cut through the hype, and the fundamental legal challenge posed by AI boils down to this: traditional laws are built on the assumption that a human is behind every key action – investing, creating, communicating, or making decisions. AI complicates or entirely eliminates that human role.
If a company’s chatbot incorrectly offers a discount it isn’t allowed to provide, who is responsible for that commitment? If an AI system is trained on vast numbers of copyright books without authorisation, does that amount to theft, or is it comparable to a researcher reading extensively before producing an original work? And when an AI generates a genuinely new invention, can it be credited as the inventor, or does current patent law simply lack any provision for such a scenario?
Globally, regulators and courts are responding to these issues in varied and often interim ways. In the United States, the approach relies heavily on judges resolving disputes through individual court cases, particularly in copyright matters. The European Union has pursued broad legislation via the AI Act, yet its implementation continues to face delays amid rapid technological and political changes. India is taking a characteristically flexible route- relying on advisories, intermediary obligations, and non-binding guidelines instead of a comprehensive AI law – while also developing targeted rules for AI use by judges and lawyers in court settings.
The landscape remains unsettled. What is certain, however, is that ignoring these evolving rules is know longer a viable options for businesses, developers, or legal professionals involved with AI.
Use of Legal Jargon
Before we go any further, it’s worth explaining a few important legal terms that keep coming up in AI disputes. These concepts are doing a lot of heavy lifting right now.
Fair use is a key part of U.S. copyright law. It lets people use copyrighted materials without permission if- after weighing four main factors (the purpose and character of the new use, the nature of original work, how much was taken, and the effect on the original’s market)- the use feels transformative rather than just a replacement for the original.
Authorship and inventorship are the basic requirements for getting copyright and patent protection. Traditionally, both have needed a real human being behind them. So far, courts have refused to give these rights to machines, no matter how creative or independent their output looks.
Negligent misrepresentations is a legal claim that applies when someone makes a false statement, had a duty to be careful, was careless in what they said, and caused harm to someone who reasonable trusted that statement. This works just as well for misleading comments from a chatbot as it does for a human salesperson.
Vicarious liability and agency principles decide whether a company can be held responsible for what its AI system does. In practice, the law often treats the AI as an extension of the business rather than a completely independent actor.
Safe harbour rules provide legal protection for online platforms, shielding them from liability for content generated by users- as longer as they meet certain responsibility and oversite requirements. These protections are being tightened in several places especially because of concerns over AI generated material.
Finally, mens rea- the “guilty mind” or mental state needed for criminal liability- remains a tricky, unresolved issue when the harmful output comes from an algorithm instead of a person.
The Proof
The sheer volume of lawsuits is the clearest sign yet that courts, not legislatures, a busy writing a real- world rulebook for AI right now. More than fifty copyright infringement cases against AI developers are already moving through U.S. federal courts, with observers counting roughly eighty generative- AI copyright disputes nationwide. Importantly, no appellate court has issued a decision on the core issues, which means every ruling so far from district courts is only persuasive, not binding precedent. And the money involved is no longer theoretically: one settlement in this wave hit $1.5 billion- reportedly the largest copyright recovery in U.S. history- covering about half a million works.
The regulatory picture shows a similar story of a system still finding its footing. The EU’s AI Acts came into force in August 2024 with a staggered rollout, but by mid- 2026 European lawmakers had already pushed back the key compliance deadline for high- risk AI systems by sixteen months (from August 2026 to December 2027), admitting that the necessary technical standards simply weren’t ready. India has chosen a lighter touch so far, skipping a single comprehensive AI law. Instead, it updated its existing IT Rules in February 2026 to require clear, ongoing labelling of AI-generated content and faster takedown timelines foe platforms. A dedicated Artificial Intelligence (Ethics and Accountability) Bill remains stuck as a private member’s bill and hasn’t become law. Even India’s judiciary has had to adapt quickly: in June 2026, the Supreme Court’s AI Committee released draft guidelines on how AI tools can be used inside court proceedings- a telling example of the legal system trying to figure out how to govern its own use of the technology.
Case Laws
- In Thaler v. Vidal (Fed. Cir. 2022) and the related UK case Thaler v. Comptroller- General of Patents [2023] UKSC, Dr. Stephen Thaler asked for patents listing his AI system DABUS as the sole inventor. Both the U.S. Federal Circuit and the UK Supreme Court turned him down, making it clear that patent law requires a human inventor. Just owing the AI doesn’t make you the inventor. Just owing the AI doesn’t make you the inventor. Patent offices in Australia, New Zealand, and Israel reached the same conclusion, though South Africa was an outlier and granted a DABUS patent without really digging into the inventorship question.
- In Thaler v. Perlmutter- Thaler also tried to copyright an image created entirely by DABUS, but that effort failed too. The D.C. Circuit ruled that human authorship is still a fundamental requirement for copyright. The U.S. Supreme court declined to take up the appeal in March 2026, so that principle stands for now. A separate case looking at how much prompting and editing is needed to qualify for protection is still working its way through the courts.
- In Bartz v. Anthropic PBC (N.D. Cal.), authors sued over the use of their books to train Claude models. Judge William Alsup refused to say all AI training is automatically fair use. Instead, he drew a practical distinction: training on lawfully purchased and digitized copies counted as fair use, but keeping pirated copies did not. The loss on the piracy issue helped push a $1.5 billion settlement, while the fair use ruling on legitimate material has become one of the most cited decisions so far.
- Thomson Reuters v. Ross Intelligence (D. Del.) looked at a legal research AI tool trained on Westlaw’s copyrighted headnotes and key numbers. The court ruled in favour of Thomson Reuters on summary judgment, finding the material was original and protected, and that using it to train a competing product wasn’t fair use- especially because it acted as a commercial substitute. The case is now on appeal to the Third Circuit and is expected to give one of the first appellate-level insights into AI training and fair use.
- The New York Times v. OpenAI and Microsoft (S.D.N.Y.), filed back in December 2023, is still in the discovery phase. The Times claims ChatGPT was trained on its copyrighted articles and can spit out close reproductions. A January 2026 court order forcing OpenAI to hand over a big batch of anonymized conversation logs was a big win for the newspaper, showing that many of these cases will hinge on hard evidence as much as legal theory.
Conclusion
None of these rulings amount to a full-blown “AI law” just yet. What they do show is judges thoughtfully applying well-established legal principles- like fair use, the need for human authorship, rules against negligent misrepresentation, and basic professional honestly- to a powerful new technology that didn’t even exist when those rules were written.
So far, these traditional legal tools have held up surprisingly well, provided judges are willing to make careful, case-by-case distinctions instead of handing down blunt, one-size-fits-all decisions.
For example, training AI on lawfully obtained material can qualify as fair use, but using pirated copies probably won’t. A machine can create something inventive, but it still can’t be named the legal inventor. And a chatbot can absolutely commit a company to a contract, no matter how much the company tries to treat it as some independent entity.
Meanwhile, lawmakers are moving cautiously. The EU is already fine-tuning its landmark AI Act even before most of its rules take full effect, while India is preferring targeted tweaks to existing laws and guidelines rather than rushing into one giant new regulation.
For companies building or using AI- and for the rest of us who rely on its outputs- the message is clear and practical: keep good records of where your training data came from, always double-check what the AI produces, follow any labelling requirements, and never forget that the person or organisation deploying the tool is the one who’ll ultimately be held responsible in court.
FAQs
Q1. Can an AI system own copyright or be named as a patent inventor?
No. courts and patent offices worldwide- from the US and UK to Australia and beyond- have consistently ruled that only real human beings (natural persons) can be authors or inventors. AI can be a fantastic assistant, but the human using it must supply the key creative ideas or expression for any legal protection to kick in.
Q2. Is it legal to train AI model on copyrighted books or articles without permission?
It really depends on how the material was obtained and what you do with it. In the US, courts have largely accepted that training on legally acquired copies can qualify as fair use. But working with pirated copies usually crosses the line. Several big cases on this topic are still working their way through the courts, so the rules are still settling.
Q3. Who is liable if a company’s AI chatbot gives a customer wrong information?
The company behind the chatbot is generally on the hook- just as if one of its human employees had slipped up. Judges have pushed back on the idea that the AI is some separate “entity” that lets the company off the hook (for example, in the Moffatt v. Air Canada case).
Q4. What if a lawyer or professional relies on inaccurate AI-generate content?
The usual professional standards of care and accuracy still apply- no excuses just because “the AI got it wrong.” In the well-known Mata v. Avianca case, lawyers were sanctioned for filling fake citations hallucinated by AI, and courts elsewhere have sent the same message: you must always check the output yourself.
