{"id":7402,"date":"2026-06-11T07:11:11","date_gmt":"2026-06-11T07:11:11","guid":{"rendered":"https:\/\/dexpatent.com\/insights\/?p=7402"},"modified":"2026-06-11T07:12:24","modified_gmt":"2026-06-11T07:12:24","slug":"ai-created-innovations-intellectual-property-rights","status":"publish","type":"post","link":"https:\/\/dexpatent.com\/insights\/2026\/06\/11\/ai-created-innovations-intellectual-property-rights\/","title":{"rendered":"Who Owns AI-Created Innovations? Understanding AI and Intellectual Property Rights"},"content":{"rendered":"

\"AI<\/p>\n

 <\/p>\n

Artificial intelligence (AI) is no longer a distant \u201cfuturistic\u201d idea, it is now embedded in labs, design studios, and software pipelines, actively generating new inventions, creative works, and functional code. This shift forces a critical question for inventors, R&D teams, and business owners:\u00a0who owns the intellectual property (IP)<\/a> when AI participates in or even drives the creative or inventive process?<\/p>\n

From patents on AI\u2011assisted drug\u2011discovery and machine\u2011learning\u2011based control systems, to copyright in AI\u2011generated images, music, and text, the traditional IP frameworks are being strained. Core concepts such as inventorship, authorship, originality, and technical effect are being re\u2011examined in courtrooms, patent offices, and policy debates. In some jurisdictions, the rules are being reinterpreted; in others, they are being quietly updated through guidelines and administrative practice. For anyone building or using AI\u2011enabled innovations, understanding how IP law adapts to AI is no longer optional, it is a strategic necessity.<\/p>\n

How AI intersects with patents<\/strong><\/h2>\n

When an AI\u2011enabled system helps design a new drug molecule, optimise a semiconductor layout, or configure a manufacturing\u2011control algorithm, the output is typically a\u00a0technical solution to a technical problem. Under classic patent doctrine, what really matters is whether the invention is\u00a0novel,\u00a0non\u2011obvious, and\u00a0industrially applicable, not who or what \u201ccame up with\u201d the idea. However, the\u00a0formal requirement\u00a0that an inventor must be a natural person has created a doctrinal tangle at the heart of modern AI\u2011driven innovation.<\/p>\n

Consider the now\u2011famous\u00a0DABUS case, where an AI system named DABUS was listed as the sole inventor in patent applications for a food\u2011storage device based on beverage\u2011container geometry and a warning\u2011light\u2011based emergency signal. Patent offices in the US, UK, Europe, and South Korea all rejected those applications, insisting that an \u201cinventor\u201d must be a human being. Courts and offices have generally held that the\u00a0user<\/em>\u00a0or\u00a0developer\u00a0of the AI, the person who framed the problem, set the parameters, selected the training data, and interpreted the AI\u2019s outputs should be treated as the legal inventor, not the AI itself.<\/p>\n

AI\u2011related inventions and subject\u2011matter eligibility<\/strong><\/h2>\n

Beyond inventorship, a second layer of difficulty is\u00a0subject\u2011matter eligibility. Many jurisdictions historically exclude \u201cabstract schemes\u201d or \u201crules and methods of mental activity\u201d (for example, pure mathematical formulas or business\u2011logic algorithms) from patent protection.<\/p>\n

\"AI<\/p>\n

Modern AI\u2011related inventions, however, often sit at the boundary: consider a neural network-based power management algorithm that dynamically tunes a smartphone\u2019s CPU and screen to maximise battery life. If the claim is framed as \u201ca neural network that learns user behaviour,\u201d it may look like a mere algorithm. But if the specification clearly links the AI module to concrete hardware (battery, sensors, thermal management circuitry) and measurable technical effects (reduced power draw, lower heating, longer uptime), the same invention can be treated as a\u00a0technical solution\u00a0and thus patent\u2011eligible.<\/p>\n

China and the European Patent Office (EPO)<\/a>, for instance, now emphasise that AI\u2011related inventions must solve a technical problem using technical means and demonstrate a technical effect (for example, faster processing, fewer errors, better resource utilisation). Merely saying \u201cwe use machine learning\u201d without tying it to such effects is unlikely to pass muster.<\/p>\n

AI and Patent Protection in India<\/strong><\/h2>\n

India currently follows the principle that an inventor must be a natural person. While the Indian Patents Act, 1970 does not expressly address AI inventorship, patent applications require the identification of human inventors. Further, Section 3(k) excludes mathematical methods, business methods, algorithms, and computer programs per se from patentability. Accordingly, applicants seeking protection for AI-related inventions should focus on demonstrating a technical application, technical effect, and practical implementation rather than merely claiming an AI model or algorithm. Careful drafting that highlights the interaction between AI and technical systems can significantly improve patentability prospects in India.<\/p>\n

Copyright and AI\u2011generated content<\/strong><\/h2>\n

While patents focus on inventions, copyright is tripping over AI\u2011generated text, images, and music. The US Copyright Office has held that works produced solely by AI, without meaningful human authorship, are not protectable under copyright, because the US system requires a human author. By contrast, some other jurisdictions are more flexible in practice: the Beijing Internet Court has described generative\u2011AI output as the result of human intellectual investment when a person selects prompts, curates results, and integrates them into a creative work.<\/p>\n

\"Copyright<\/p>\n

At the same time, generative AI platforms face infringement allegations when they are trained on copyrighted works without explicit permission. In the US, several authors and artists have sued major AI companies, arguing that their books and artworks were used in training data, effectively creating unlicensed derivatives.<\/p>\n

At the same time, another emerging issue concerns the datasets used to train AI models. Many generative AI systems are trained on vast quantities of text, images, music, software code, and other copyrighted materials. This has led to ongoing legal disputes regarding whether the use of such materials for AI training constitutes copyright infringement, fair use, or a separate category of lawful activity. As AI adoption grows, the ownership and licensing of training data are expected to become increasingly important aspects of intellectual property strategy<\/a>.<\/p>\n

What this means for you as an inventor or business<\/strong><\/h2>\n

If you are using AI to:<\/p>\n