The explosion of generative artificial intelligence tools is transforming industries by enabling the rapid creation of text, images, music, code, and more. As businesses increasingly deploy generative AI to streamline content creation, design new products, or even develop software, they must grapple with the legal implications of such outputs. Central to this conversation are the questions around ownership, licensing, and the treatment of derivative works generated through AI technologies.
At the heart of these issues lies intellectual property (IP) law, which was largely conceived in a pre-AI era. Generative AI challenges fundamental concepts such as authorship, originality, and rights to derivative creations. Businesses using AI in their operations face a complex legal matrix involving both input data (training sets) and output (the generated content). Navigating this landscape responsibly is crucial for mitigating legal risks, protecting proprietary works, and ensuring ethical deployment of these powerful tools.
The Nature of Derivative Works in AI-Generated Content
A derivative work, under copyright law, is generally understood to be a new creation that is based upon one or more pre-existing works. Examples include translations, adaptations, or any material that recasts the original in a new form. This concept becomes murky in the context of AI when the output resembles or is inspired by the structure, style, or content of the material used to train AI models.
Generative AI models are typically trained on vast datasets that may include copyrighted texts, images, music, code, and more. These datasets often span the open web, including both licensed and unlicensed sources. While some developers assert that training on public data qualifies as fair use, this is far from a settled legal norm, and lawsuits across various jurisdictions are poised to clarify the boundaries.
Where outputs are similar in tone, structure, or composition to training materials, the risk arises that they could be interpreted as derivative works. For example, an AI that generates a new image in the distinctive style of a particular living artist could produce a work that infringes on the original artist’s moral rights or copyright, depending on applicable national laws.
The Question of Ownership Over AI-Generated Outputs
Perhaps the most contested question around AI-generated content is who—if anyone—owns the copyright. Traditional copyright laws typically require human authorship. That is, for a work to be protected, it must be the result of human intellectual effort. This presents a dilemma when outputs are autonomously generated by AI with minimal human intervention.
In the United Kingdom, for instance, the Copyright, Designs and Patents Act provides that the author of a computer-generated work is “the person by whom the arrangements necessary for the creation of the work are undertaken.” This offers a limited form of protection, suggesting that the entity initiating or prompting the AI could hold copyright. However, such protection is weaker and may not be internationally recognised, creating uncertainties for works circulated across borders.
In commercial settings, a common approach is to assign ownership contractually. Companies may stipulate in employment or vendor agreements that all AI-generated content developed under contract, irrespective of the specific authorship mechanisms, will be treated as ‘works made for hire’ or otherwise assigned to the commissioning party. While this can offer internal clarity, it does not eliminate the potential for disputes about third-party rights, especially when training data was pulled from unclear or unauthorised sources.
Source Material and the Importance of Data Licensing
One of the foundational challenges in the generative AI value chain is the provenance of the data used to train or fine-tune AI models. Commercially deploying AI models trained on copyright-protected materials without proper licensing could expose companies to infringement claims. Lawsuits are already emerging in this space—from class-action litigations in the United States involving visual artists and coders, to regulatory inquiries into how data from books, news articles, and academic papers has been repurposed by AI developers.
For businesses relying on third-party AI services, due diligence around the scope and terms of data use is crucial. Licensing agreements should be scrutinised to ensure that any output generated from the service does not encroach on restricted content, and that the vendor has legal rights over the training data used in its models. This issue is particularly salient when using APIs from major AI providers, some of whom offer different tiers of output licensing depending on payment levels and end-use purposes.
Open-source models present a parallel concern. Many organisations are drawn to open-source AI for its flexibility and cost-effectiveness. However, the licence terms attached to these models vary widely. Some, such as those under more permissive licences, may permit commercial use without attribution. Others are bound by non-commercial clauses or share-alike provisions that restrict how derivative outputs can be modified or monetised. Proper interpretation of these terms—as well as confirming any embedded datasets used in model training—is a fundamental part of responsible AI governance.
Contractual Safeguards in AI Deployment
Given the significant ambiguity in laws relating to AI outputs, contracts become key instruments for managing risk. Businesses deploying or commercialising generative AI should pay close attention to their service-level agreements, vendor contracts, and end-user terms.
Central to these contracts should be clarifying ownership of generated outputs. Where third-party AI services are used, businesses should ensure that service agreements grant them explicit rights to use, adapt, distribute, and potentially commercialise AI-generated content. Equally important is confirmation that the vendor provides indemnity for claims arising from unauthorised use of data in model training or output generation.
Internal governance policies may also require employees to adhere to a framework that ensures the content generated conforms with company standards, respects third-party rights, and avoids sensitive material. As AI tools are increasingly integrated into marketing, product design, and software development, having a clear internal policy for usage, review, and documentation of outputs can curb the risk of infringing on others’ IP rights or breaching contractual obligations.
The Layer of Moral Rights and Style Mimicry
Beyond textbook copyright issues, the notion of moral rights plays a significant role, particularly in jurisdictions such as France, Germany, and many Commonwealth countries. Moral rights refer to an author’s right to be credited for their work and to object to derogatory treatments that could harm their honour or reputation.
In generative AI, these rights can become implicated when models generate content that mimics or parodies the style of living artists or authors. An AI being prompted to generate paintings in the style of Banksy or write poetry mimicking Carol Ann Duffy may flirt with moral rights infringement, particularly if there is a risk the public might confuse the provenance of the output. If a business uses such mimicked content in advertisements or monetisable products, moral rights claims—even without direct economic infringement—could lead to legal challenges or reputational backlash.
While moral rights cannot always be contractually waived or bought out—especially in countries with stringent protection regimes—they do demand that businesses consider ethical and reputational dimensions of their AI-generated outputs beyond questions of economic copyright ownership.
Licensing Models for AI-Based Products
As AI outputs gain value in the marketplace, productising them requires clear licensing structures. Businesses may need to consider how to license AI-generated content for end users—whether content is resold, integrated into services, or transformed into consumer experiences such as games or design assets.
In developing a licensing approach, businesses should distinguish between raw AI output and value-added creations. For example, an AI-generated image may be treated as a raw file, but the curated, edited, and framed version of it embedded in a digital product could constitute a new original work. Moreover, tools that allow end-users to prompt AI systems to generate content must consider how final outputs are licensed to the user—does the user own the content they prompt, or does the platform retain the rights?
Some AI service providers offer ‘custom licensing models,’ which grant clients exclusive commercial rights to outputs under certain subscription tiers. Others maintain joint rights over outputs or reserve the right to reuse outputs for further training. Understanding these distinctions—and ensuring they align with your business model—is critical for downstream monetisation and IP compliance.
Cross-Border Considerations and Jurisdictional Diversity
Another layer of complexity stems from the global nature of AI development and deployment. IP laws vary significantly across jurisdictions. While UK law permits rights in computer-generated works, the US until recently required human authorship for copyright protection. The US Copyright Office has taken a strong stance that purely AI-generated works are ineligible for protection unless there is significant human involvement in the expressive choices of the output.
For businesses operating transnationally, this patchwork of legal standards complicates decisions on licensing, content deployment, and enforcement of rights. An AI-generated advertising video that is protected in one country might be considered public domain in another. This is particularly daunting in online marketplaces, where digital content flows readily across borders.
To manage this, businesses can adopt ‘highest standard compliance’ policies—ensuring that whatever content is shared globally meets the strictest applicable standard in its likely markets. Alternatively, AI governance teams may maintain content segregation, deploying different versions for different regions based on their legal frameworks.
Dispute Resolution and Legal Precedents
As AI-generated content becomes more commercially relevant, legal cases will increasingly define the contours of rights and obligations. Emerging disputes—between artists and tech companies, coders and platform providers, and businesses versus regulators—will help shape what constitutes fair use, infringement, and authorship.
This shifting jurisprudence highlights the importance of dispute resolution mechanisms in AI-related contracts. Clear agreements should identify governing law, jurisdiction, and whether arbitration or litigation will resolve conflicts. For companies employing user-generated prompts—like AI writing or image generation platforms—terms of service should include language limiting liability, disclaiming authorship, and establishing usage rights.
Companies with robust AI footprints should also monitor regulatory developments. Initiatives in the EU, such as the Artificial Intelligence Act, and ongoing consultations in the UK and US, will likely lead to sector-specific guidance on IP. Engaging with legal counsel familiar with both AI technology and copyright norms is increasingly a necessity, rather than a luxury.
Ethical and Social Implications of Derivative AI Creations
Beyond technical compliance, the social impact of creating AI content derived from other people’s styles, likenesses, or work cannot be ignored. For example, deepfakes or synthetic voices mimicking celebrities may comply with contractual terms, but still provoke ethical backlash or concern about misinformation and consent.
Companies should consider embedding ethical frameworks into their AI governance. This may include checklists for data use consent, policies on impersonation, respect for deceased individuals’ legacies, and community advisory boards for sensitive content. Public trust is becoming a valuable currency in the AI space, and how companies handle derivative works through an ethical lens will influence brand perception.
Conclusion
Operating a generative AI business today is both a legal and ethical balancing act. Derivative works and licensing introduce significant complexities that touch upon copyright, moral rights, contractual law, and international legal variability. With legislation struggling to keep pace with innovation, it falls upon businesses to adopt diligent, proactive strategies grounded in legal prudence and ethical responsibility.
Those who succeed in navigating these waters will be able to confidently commercialise AI-generated content while mitigating risk and fostering trust. Whether through contracts that assign clear output ownership, rigorous data licensing audits, or thoughtful policies on style emulation and user interaction, businesses can set safeguards that secure both innovation and compliance. Ultimately, handling derivative works and licensing in generative AI is not just a legal necessity—it’s a sign of responsible leadership in the digital age.