Legal IP Issues in Generative AI

The emergence of generative artificial intelligence has fundamentally transformed the landscape of content creation. Tools powered by large language models, diffusion models, and other machine learning architectures are now capable of producing music, visual art, written text, software code, and even video at remarkable speed and scale. From OpenAI's ChatGPT to image generators like Midjourney and Stable Diffusion, these systems have moved from research laboratories into mainstream commercial use within a remarkably short period. Yet their rapid adoption has outpaced the development of legal frameworks needed to address the profound intellectual property questions they raise.

For businesses, creators, and legal practitioners in Malaysia and beyond, understanding the IP implications of generative AI is no longer optional. The questions are not merely academic: they affect how companies deploy AI tools, how creators protect their work, and how courts will adjudicate disputes that are already making their way through legal systems worldwide.

The Ownership and Authorship Problem

Traditional creative works have a clear chain of authorship. A painter creates a painting; a novelist writes a book; a composer produces a musical score. Copyright law in virtually every jurisdiction is built on this premise of human authorship, granting exclusive rights to the person who created the original work. Generative AI disrupts this foundational assumption in ways that existing legal frameworks were never designed to address.

When a user inputs a text prompt into an AI image generator and receives a detailed illustration in return, the question of authorship becomes deeply contested. Is the author the programmer who designed and trained the AI model? Is it the user who crafted the prompt that guided the output? Or could the AI system itself be considered the author? Each answer carries dramatically different legal consequences for ownership, licensing, and the ability to enforce copyright protection.

In most jurisdictions, including the United States, the prevailing legal position is that copyright protection requires human authorship. The US Copyright Office has consistently refused to register works created entirely by AI without meaningful human creative input, as demonstrated in its 2023 guidance on AI-generated works. This position was reinforced in the case of Thaler v. Perlmutter, where the court upheld the Copyright Office's refusal to register an AI-generated artwork, affirming that human authorship is an essential requirement for copyright protection.

Malaysia's Copyright Framework and AI-Generated Works

Malaysia's Copyright Act 1987 provides protection for original literary, musical, artistic, and other categories of works. Crucially, the Act requires that a work be original and that it involve sufficient human intervention or involvement in the creative process. The concept of originality under Malaysian law does not demand novelty in the patent sense, but it does require that the work originate from the author through the exercise of skill, labour, and judgment.

Computer-generated works occupy an uncertain position under Malaysian copyright law. While the Act does not explicitly address AI-generated content, its requirement of human involvement suggests that purely machine-generated outputs, produced without meaningful human creative contribution, may not qualify for copyright protection. This creates a significant gap in protection for businesses that rely on AI tools to generate marketing content, design elements, or other creative assets.

The practical implications are considerable. If an AI-generated logo, advertisement, or piece of written content cannot receive copyright protection, the business that commissioned or prompted the creation may have no exclusive rights over that output. Competitors could freely reproduce the same or similar content without legal consequence, undermining the commercial value of AI-assisted creative work.

Copyright Infringement and the Training Data Problem

Beyond questions of ownership, generative AI raises serious concerns about copyright infringement at the input stage. These models are trained on massive datasets that frequently include copyrighted material, from books and articles to photographs, illustrations, and music. The models learn patterns, styles, and structures from this training data, and their outputs can sometimes bear striking resemblance to specific copyrighted works in the dataset.

This has given rise to a wave of litigation that is testing the boundaries of copyright law globally. In Andersen v. Stability AI Ltd, a group of visual artists filed a class action lawsuit alleging that Stability AI, the developer of Stable Diffusion, trained its model on billions of copyrighted images scraped from the internet without authorisation, and that the model's outputs constitute derivative works that infringe their copyrights. Similarly, in Getty Images v. Stability AI, the stock photography giant alleged that Stability AI copied over 12 million images from its library without permission or compensation to train its AI model, with some outputs even reproducing a modified version of Getty's watermark.

These cases are still being litigated and are expected to produce landmark rulings that will shape the legal framework for AI development for years to come. The outcomes will determine whether the use of copyrighted material for AI training constitutes infringement or falls within permissible boundaries.

The Fair Use Defence and Its Limits

Defendants in AI copyright cases are widely expected to invoke fair use and similar doctrines as their primary defence. In the United States, the fair use doctrine under Section 107 of the Copyright Act considers four factors: the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect on the market for the original work. AI developers argue that training constitutes a transformative use because the model learns abstract patterns rather than copying specific works, and that the outputs serve different purposes from the original training data.

However, rights holders counter that the sheer scale of copying involved in training, often billions of works, distinguishes this from traditional fair use scenarios. They also argue that AI-generated outputs directly compete with and reduce the market for human-created works, satisfying the fourth factor of the fair use analysis. The resolution of these arguments will have profound consequences for the AI industry and creative professions alike.

In Malaysia, there is no general fair use provision equivalent to the US doctrine. Instead, the Copyright Act 1987 provides specific fair dealing exceptions for purposes such as research, private study, criticism, review, and news reporting. Whether the use of copyrighted material for AI training would fall within any of these narrow exceptions remains untested in Malaysian courts, adding further uncertainty for businesses operating in this space.

The Path Forward: Collaboration and Ethical Standards

The legal challenges posed by generative AI cannot be resolved by courts alone. Lawmakers, businesses, AI developers, and creative communities must collaborate to develop ethical guidelines and legal standards that balance innovation with the protection of intellectual property rights. Several approaches are being explored internationally, including mandatory disclosure of training data sources, opt-out mechanisms for rights holders, licensing frameworks for training data, and revenue-sharing models between AI companies and creators.

For Malaysian businesses and creators, the current uncertainty underscores the importance of proactive legal planning. Companies deploying generative AI tools should carefully review the terms of service of AI platforms, understand the limitations on ownership of AI-generated outputs, maintain records of human creative input in the production process, and seek legal advice on the IP implications of incorporating AI-generated content into their products and marketing materials.