1. Introduction

The rapid advancement of AI continues to raise complex questions about the applicability of intellectual property (IP) laws to AI and AI-generated works.

The unprecedented pace of development of this technology is pushing enterprises towards self-governance frameworks founded on ethical considerations. IP remains one of the leading and most contentious issues in respect of AI governance. We examine some of the key trends that have emerged globally at the intersection of AI and IP below.

As a preliminary comment, overall, we expect to see governments across the world grappling with balancing strategies aimed at encouraging the development of AI and innovation while, at the same time, attempting to modernize IP and AI legal frameworks to account for AI. The pressure on jurisdictions everywhere is to do these things while addressing challenges such as cross border IP infringement involving AI systems and the proliferation around the globe of non-harmonised AI laws. Despite this, some governments, such as New Zealand, have adopted a "light-touch, proportionate, and risk-based" regulatory approach, aimed at providing reassurance and fostering AI growth in the country, with the view taken that existing legislation already provides adequate guardrails and so further regulatory intervention is necessary only to unlock innovation or address acute risks. 

AI adoption continues to grow, 2025 is likely to see a wider range of commercial applications across all sectors. In light of the IP related challenges, businesses leveraging AI technologies must strategize to navigate the evolving intellectual property landscape. As the legal framework around AI continues to develop, businesses need to ensure that they avoid copyright infringement while also effectively safeguarding their own IP assets. Developing a clear understanding of how existing laws apply to AI technology and staying updated on legal developments in will be crucial for companies seeking to innovate responsibly and protect their intellectual property in an AI-driven economy.

In 2025 and beyond, we expect to see more mainstream adoption of AI tools by rights holders and advisors. Such tools should be used to streamline manual processes and quickly identify and act on infringement risks. Key team members would benefit from access and analysis of large amounts of data, which can be used to improve their IP decision-making processes. However, while AI provides increased efficiency, human oversight remains essential.

2. Rights in input data used to train AI models and infringement of IP

Training generative AI involves using large bodies of IP-protected works in ways that may be infringing under current laws. Governments seeking to “unlock” the potential of generative AI are now more often looking to legislate to permit text and data mining (TDM) of IP-protected data in order to train AI. Some jurisdictions however have delayed acting on organisations training AI without permission or attribution of the IP-protected data which they used. 

The intellectual property in the data used to train AI models is growing as a subject of legislative discourse and is now a key issue in matters that have flooded courts across the world.

It is a highly debated topic - whether use of copyright-protected materials to train AI models infringes copyright. Training AI using personal data or protected IP provides challenges to legislators worldwide. 

A trend expected over the coming year therefore is increased regulatory scrutiny of organisations that create or use AI technologies which have been trained using (i) personal data and/or (ii) information/data protected by IP rights. Regulators worldwide are now paying greater attention to balancing the benefits of AI against concerns about personal data and the protection of IP, and this will continue in 2025 and beyond. By way of example, the new Labour government elected in 2024 in the UK pledged to bring forward specific legislation tackling AI in 2025 which is expected to provide greater clarity on these IP issues, after the previous government's proposals to introduce legislation to enable the greater use of data to train AI models were abandoned after being met with backlash from the creative sector.

Generative AI has spawned a variety of copyright lawsuits around the world. 

• In the United States, for example, authors and news publishers have sued OpenAI, Microsoft, and others over their large language models (LLMs), alleging their training processes and/or their outputs involve impermissible copying.

• Artists and a photo aggregator have sued graphic generators such as Stability AI, making similar claims. A prominent case likely to provide guidance on these issues that is currently progressing through the English court system is the Getty Images v Stability AI case. In that matter, Getty alleges copyright and database infringement as a result of Stability AI downloading and storing Getty's images onto Stability AI's server to train its system. The case also raises questions about the role of user input in generating images and whether any legal exceptions to infringement might apply.

• Music publishers allege Anthropic’s AI system routinely reproduces copyrighted lyrics.

• Computer programmers claim that OpenAI and Microsoft’s use of code posted on GitHub to train AI coding assistants is outside the authorized scope of use of that code.

• And a legal research database company contends ROSS Intelligence’s AI tool improperly copies the database’s case summaries.

None of these cases in the US has yet gone to trial as at the time of writing. In pretrial proceedings in the US, the courts have generally sought to focus the disputes on core issues of copying and fair use and the courts have dismissed numerous ancillary claims, many of them alleging technical violations of the US Digital Millennium Copyright Act (DMCA). 

We expect that many cases are likely to ultimately turn on the precise details of the training mechanisms, the degree of similarity of AI model outputs to copyrighted works, and the fair use factors. 

In many jurisdictions, copyright may be infringed where a person carries out any of the restricted acts in relation to the whole or a substantial part of a work. In some cases, if the data attracts copyright protection, training an AI platform using the data is likely to amount to copyright infringement.

Many jurisdictions do not have a fair dealing exception with respect to such copyright infringement although some, such as the UK and Australia, provide exceptions for particular activities. 

• In the UK, for example, there is an exception for text and data mining (TDM), although its application is limited. In some countries there is no TDM exception and the comparatively restrictive fair dealing exceptions are unlikely to provide a safe harbour. The same protection may not be afforded to confidential data or trade secrets.

• By contrast, the Hong Kong Government recently commenced consultation for introducing a specific copyright exception for the reasonable use of copyrighted works in data mining and AI model training.

• In the US, fair use is a complicated and flexible legal concept, where courts analyze the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect of the use on the potential market or value of the copyrighted work. As with several past technical revolutions, the US Supreme Court may eventually step in with a broad pronouncement on fair use in generative AI.

• Evidence of a trend to facilitate TDM is growing – Singapore, for example, amended its copyright laws recently to address complex infringement issues, including those arising from the design and implementation of machine learning models. Amongst the key amendments were the introduction of a new exception to copyright infringement for the purpose of “computational data analysis”. This exception permits the use of large database of materials as training input for machine learning models without needing to seek permission from each rights owner. This provision is broadly similar to TDM exceptions in other jurisdictions but wider in scope, as it (i) applies for both commercial as well as non-commercial purposes; and (ii) cannot be excluded or restricted by contract. It remains to be seen how the courts will apply it in the context of AI.

However, key issues still persist, particularly with regard to copyright ownership of AI-generated works.

3. Rights in output data - AI-generated works, AI-inventions and other AI-outputs and infringement

Generative AI (that is, AI which generates text, images, speech, video or even technical inventions based on user-inputted instructions) continues to increase in capability and grow in adoption. 

Generally, IP laws are designed to protect human creations. Most copyright and patent laws, for example, do not yet explicitly address AI's role in authorship or inventorship, leaving a legal void requiring attention. 

If AI is responsible for content creation autonomously without any human input, then the question is who owns the copyright protecting such content. Traditionally, this is the person or organisation that creates the works.

• In some countries, such as the UK, this may be answered by the fact that computer-generated works will be owned by the person who made the necessary arrangements for the creation of the work.
• For Europe, the overwhelming view is that AI cannot be a legitimate author. However, specific ways of using AI may result in a work that is protected for the user.
• The Hong Kong Government recently affirmed that Hong Kong’s current copyright framework is capable of protecting works produced by generative AI and such works would receive the same protection as traditional human authored works.

• However, questions remain elsewhere: who is this person? Is it the person who created the software; or the person(s) who taught the software, or both?

• We expect to see significant developments and change globally on this issue.

A further concern is how to regulate instances where AI-generated works feature IP-protected content, such as a company logo or even the image of a real person, which could raise privacy/ data protection issues.

Until these issues are addressed — either through case law or legislative reform—uncertainty will continue to surround the subsistence of copyright, ownership, and potential infringement related to AI models and AI-generated outputs.

(a) Copyright:

The issues relating to copyright generally do not relate to the AI program itself. AI programs usually qualify as IP (software or computer programs being literary works). In some countries however, copyright protection will not apply for functional aspects of AI such as algorithms or system designs.

AI systems function however by processing human-provided instructions to generate problem-solving outcomes. This capability makes AI-based programs highly valuable from an IP perspective, as their innovative nature and diverse utility underline their significance of IP protection. If an AI operates autonomously without human intervention, ownership becomes more complex and nuanced. 

Copyright laws often require that there must be a natural person to whom copyright can be attributed and many jurisdictions do not provide for “computer generated” works where no human author is involved. This creates a gap in the protection of AI-generated works, which are typically produced autonomously with little or no human intervention.

Many copyright laws also require that “sufficient effort” must be expended to make any literary, musical, or artistic work original in character – which involves time, human labor, and skill. What constitutes sufficient effort for AI-generated content remains largely untested, raising debates about whether crafting prompts or editing AI output meets the thresholds. Additionally, if non-human entities are recognized as “authors” then copyright duration may become complex. Generally, copyright protection is granted for an author’s lifetime plus a period of time following, potentially leading to indefinite protection for AI-generated works.

The European Union’s standalone sui generis database rights are not widely replicated in similar IP laws around the world and, as such, databases and datasets may only be protected under copyright laws to the extent that the selection or arrangement of the data meets the requisite level of originality. This does not align well with the nature of AI as the type of creativity that copyright law seeks to protect does not easily apply to how modern databases are structured or how data is gathered. For instance, machine-generated ‘big data’ databases are typically compiled in a systematic manner rather than through creative selection.

A final comment relating to copyright is to note that the moral rights of a human author (such as the right of attribution) under copyright laws have not been addressed specifically for AI-creators. AI cannot decide on the fate of the work it has created per se and cannot be a subject of civil rights. Such restrictions may be another obstacle to the use of copyright by AI. Having said that, we note that, in Kazakhstan, a copyright certificate has been issued for a graphic novel created using AI as a composite work where the protection was granted for the arrangement of materials as a result of human creative activity, but not for the originality of images generated by AI.

(b) Patents:

Patents and AI-generated inventions present similar challenges. Many patent laws require the inventor to be a natural person – a requirement which could exclude AI from being independently recognized as an inventor. AI-driven innovations, such as those involving algorithms and machine learning processes, face challenges in meeting the criteria for protection as an invention. Under US patent law for example, absent at least one human inventor, an invention is not patentable.

Also, the criteria for patentability usually involves requirements for novelty, an inventive step and industrial applicability. These criteria raise questions about whether AI-generated inventions can ever meet the inventive step requirement, traditionally linked to human ingenuity. AI relies on algorithms and datasets to mimic human cognitive functions, enabling it to generate patentable inventions. 

It is expected that further cases dealing with AI-generated inventions, exemplified by the DABUS cases in multiple jurisdictions, will spark global discussions about whether AI can be recognized as an inventor and will be run in more courts seeking to challenging whether creators need to be humans and examining the issues regarding AI assisted inventions which may not have a human inventor.

The eligibility of AI to be registered as a patent holder has become an area of discourse in the legal landscape. This discourse gained traction when the South Africa Commission for Intellectual Property and Companies (CIPC) became the first global office to approve a patent listing AI as an inventor. 

Conversely, patent offices in various other regions, including the United States, India, Europe, Australia and Taiwan, have rejected similar applications, primarily emphasizing that AI lacks legal personality.

We expect to see further guidance develop around the world – similar to the 2024 US Patent and Trademark Office issued Inventorship Guidance for AI-Assisted Inventions (which provides instructions for examiners and stakeholders in determining whether inventions developed using artificial intelligence include sufficient human contributions to qualify for patent protection). This guidance acknowledges the growing use of AI in developing technology and recognizes that AI assisted inventions may be patentable and provides that inventions developed with the assistance of AI must be analyzed to determine whether the human named as the inventor made a significant enough contribution to the invention. 

Similarly, we expect to see a continuing updating of guidance as to the level and type of human contributions that are necessary to support patentability as new cases make their way through courts.

Some jurisdictions have however pressed ahead with updating their patent laws – and we expect this will accelerate. Indonesia, for example, recently amended their patent law to provide protection for certain inventions in technology that solve specific problems such that, while software alone is not patentable, AI programs that exhibit technical effects, functional characteristics, and problem-solving capabilities can be patented. 

The divergence in approach has caused debate among legal professionals, both academic and practitioners, about the formative patent process which gave rise to the decision of the CIPC. Notably, South Africa's patent approval system follows a formative process which considers the fulfilment of procedural formalities without requiring in-depth evaluations of the prospective patent's substantive merits. Other jurisdictions, such as India, the United States of America and the United Kingdom, apply a substantive examination approach which evaluates the novelty of a patent among other requirements. While South Africa's patent approval process aims to streamline the application procedure, it has sparked concerns among legal experts regarding its possible implications. In particular, the decision to list AI as an inventor has faced significant criticism, with many suggesting that the CIPC should have denied the application. 

Expect patent protection for AI-driven inventions to continue to present novel issues in the next few years.

4. AI and Digital Replicas (Deepfakes) and IP

A further trend is the increasing focus on the legal implications of deepfakes. As AI technologies continue to evolve, the rise of deepfakes is one of the more concerning developments.

A recent report commissioned by Mastercard in New Zealand illustrated the trend - nearly 29% of individuals and 18% of businesses were targeted by deepfake scams in the past year and 47% of them fell for the manipulated content. This trend, further fueled by advancements in generative AI, is likely to continue escalating.

AI models are becoming more intelligent, requiring fewer source materials to create convincing deepfakes. This means that even individuals with a small online presence could be at risk. A handful of publicly available images or audio samples could soon be enough for someone to create a realistic deepfake, raising concerns for those who may not be aware of how much of their personal information is available online. As the technology advances, the platforms are able to create manipulated videos and audio in real-time. This may pave the way for live-streamed deepfakes, further increasing the risks.

Most IP laws are ill-equipped to address the challenges posed by deepfake technology.

Copyright law generally is not fit for purpose in respect of deepfakes as the source material for many deepfakes either falls outside the scope of copyright protection or the copyright owner is not the individual who is harmed by the infringement.

We predict that governments will take more action to address harmful deepfake content. Australia has done that in 2024 by outlawing creation and non-consensual dissemination of sexuality explicit material online which is created or altered using generative AI. Expect to see similar measures introduced.

Possible causes of action presented by deepfakes include copyright infringement (if a deepfake involves unauthorized use of copyrighted material), trademark infringement (if it uses a registered trademark without permission), the tort of passing off (if it misrepresents a product or service as endorsed by a well-known individual), personal data privacy violations or defamation (if the content defames an individual).

In July 2024, for example, the US Copyright Office issued Part 1 of a report on Copyright and Artificial Intelligence, relating to Digital Replicas (deepfakes). 

• The report addresses the growing concern over protecting the rights of creators and individuals with respect to unauthorized digital replicas.

• It recommends that the new laws be enacted and remedies be provided to mitigate the harm caused by deepfakes, including privacy, unfair competition, consumer protection, and fraud.

Additional reports will likely be issued in 2025-26 regarding training data and other issues raised by the growing development and use of AI technologies.

5. Contractual frameworks relating to IP infringement and AI systems

Liability for infringement also remains an issue despite most contracts between AI system owners and end-users dealing with this topic. Many AI system owners seek to exclude or limit liability arising from provision of their services to end-users. 

The issue of whether the outputs of AI models – particularly where they substantially reproduce source materials - may infringe copyright and (if so) who may be responsible for such infringement, is also unresolved. Is the person who infringed the user of the AI generated work without the rights holder’s consent or the AI developer or AI system owner? 

This ambiguity poses risks for businesses. If there is no clear proprietary right in AI-generated works, businesses may be exposed to unnecessary risk. For instance, such works could be freely copied by competitors without infringing. It may also be difficult to monetize such works (e.g. by licensing) where ownership of copyright in the work is unclear or lacking.

6. AI and IP Management (freedom to operate searches, infringement risk mitigation and portfolio optimisation)

(a) Management: 

The integration of AI into IP practices is expected to become more mainstream in future, freeing up advisors to provide more valuable legal support. AI technology is already being used to improve process efficiencies, for efficient data analysis and for managing trademark portfolios. AI tools are now increasingly being used for searching, reviewing and analysing IP registers, including trademarks, designs and patents for the purpose of freedom to operate analysis. In addition to IP registration, AI is also being utilised to protect intellectual property. Some companies with high-value patents, for example, are using AI to review newly registered patents, protecting their patents and avoiding duplication.

This trend is expected to accelerate in 2025 and onward as the AI tools improve and are further developed. We expect to see more bespoke offerings for monitoring and enforcement. Nevertheless, such AI tools are still in developmental stages and often still provide an imperfect initial due diligence which requires subsequent review by a human analyst.

(b) Portfolio strategy, optimisation and protecting IP:

AI tools are streamlining how IP portfolios are managed, and key decisions are made.

Some key trends include:

• Automation of routine tasks: AI tools can be used to automate routine portfolio reviews and analyse data more effectively, leading to faster decision making about new filings and renewals. While AI has been used for several years in global freedom to operate searches, we anticipate the development of specialised tools that focus on reviewing search data and providing actionable insights from the results. This will save time for advisors and improve the accuracy of their advice.

• Prediction and analysis: AI can analyse vast amounts of data in seconds, making it an ideal tool for predicting market trends, identifying potential infringements, and assessing the viability of new brands. For example, AI could start making smart suggestions about a trademark portfolio based on emerging industry trends or portfolio gaps.

• Unparalleled marketplace monitoring: Companies are already using AI tools to efficiently monitor activity across digital channels and markets. Gaining access to this data means IP teams can communicate and act on infringement risks easily and efficiently. We expect to see more mainstream adoption of such tools.

• Efficient rebranding: AI tools can be used to identify and remove outdated digital marketing content such as old logos from websites, social media platforms and marketplaces. These tools can help prevent consumer confusion as well as free up time to monitor more serious infringements.

• Specification drafting: While open AI tools can be a useful starting point for drafting specifications, we anticipate the development of more specialised AI tools that are trained on up-to-date data from available IP offices. Ideally, these tools would harmonise IP policies across key jurisdictions and offer intelligent suggestions for drafting trademark and patent specifications, aiming to minimise avoidable formality objections. However, the effectiveness of these tools will be limited by the quality of record keeping at each IP office, meaning human oversight will remain essential.

(c) AI use in enforcement of IP rights

We expect to see more adoption of AI tools in the enforcement space, particularly for detecting and dealing with infringements.

• Detecting infringements: The availability of generative AI has enabled bad actors to create convincing infringing material at unprecedented speed and in great volume. By using image recognition and machine learning, AI can scan websites, online marketplaces and social media for counterfeit products, fraudulent websites, and scams. Some tools can analyse large volumes of digital content to identify infringing material and provide a risk assessment. These tools look for red flags such as spelling mistakes, domain redirection and other signature signs. By analysing these signs, the tools can flag high risk content, helping brands respond quickly without the need to manually review each result.

• Smarter detection: We expect AI tools will become more efficient at detecting infringing content as they are able to learn from previous findings. Third party systems should be in a position to train their models on up-to-date infringement trends, meaning rightsholders will benefit from data on current scams and infringement trends, improving their detection capabilities and usefulness for brands.

• Automated infringement notices: AI tools can be used to automatically generate takedown notices, meaning brands can accelerate the process of removing infringing content from online platforms. This means rightsholders can take immediate action to safeguard their reputation and prevent further harm to their consumers.

We expect to see brands making more use of these tools to streamline their monitoring processes.

(d) IP Litigation and AI

• Document review and summary: Legal practitioners are already using sophisticated AI tools for reviewing and summarising large volumes of documents. AI tools can quickly scan and summarize large quantities of evidence, identifying the most relevant information for the task at hand. This allows lawyers to focus more on strategy, rather than time consuming document review. However, as with all AI platforms, some level of human oversight is essential.

• Streamlining evidence preparation: Tasks like preparing evidence of use, which are essential in trademark prosecution and litigation, can also be significantly streamlined by the use of AI tools. By automating evidence collection and organisation, AI tools can save valuable time for lawyers and clients alike.

7. Conclusion

In conclusion:

• The intellectual property in the data used to train AI models has been the subject to legislative discourse and a key issue in matters that have flooded courts across the world.

• A highly debated topic is whether use of copyright-protected materials to train AI models infringes copyright. Training AI using personal data or protected IP continues to provide a challenge to legislators worldwide. 

• In most countries, authorship of creative works and invention of new technology can only be attributed to humans and can be procured by corporations via a work-for-hire arrangement.

• A vital question that arises is whether AI can be regarded as a legitimate author of content that it generates or an inventor in case of patents, given the want of legal personality of the AI itself.

• Pertinently however, the Commission for Intellectual Property and Companies in Africa, was the first global office to have granted a patent application where AI was the inventor. This move had received considerable backlash from other countries.

• However, the Hong Kong Govt. has declared AI generated works as being capable of copyright protection under the existing law.

• The US Patent Office has also issued a nuanced Inventorship Guidance providing a framework for examiners of patent applications to assess the quantum of human contribution for the invention to qualify for patent protection – a move seeking to balance IP rights with the need to leverage upcoming technology.

• In the UK as well, the law specifically permits copyright protection in “computer-generated works” though the broader question of originality being a precondition for IP protection continues to be ambiguous.

• Consequent to the ambiguity around copyright ownership over AI generated content, the issue of IP infringement as a result of AI has taken center-stage in global legislative discourse.
  • AI models are trained on IP-protected data.
  • While in some cases the right to scrape has been set out contractually between AI models and end-users, many large enterprises like Microsoft’s OpenAI have been sued in various countries, in respect of unauthorized scraping of copyrighted work resulting in nuanced questions around fair use through democratized data mining of works in public domain pending decision before the judiciary. Both issues have resulted in the large AI companies and AI based music and video generators being sued in courts across jurisdictions. Interestingly, many governments are leveraging AI to detect infringement and mitigate the risks. In a landmark judgment of the Hamburg Regional Court, it was held that even a machine readable disclaimer of a website specifically precluding the scraping thereof for the purpose of data mining would not preclude such mining done for scientific research that was publicly available without a cost.
  • Further, with AI having the capability to generate images with unprecedented quality, it has brought to the fore a unique issue within the larger ambit of infringement – deepfakes. Thus far, legislations have recognized impersonation as an offence under penal, privacy and information technology laws. However, in a landmark development, courts have recognized the personality of celebrities as being monetizable assets which are prejudiced by the emergence of deepfakes along with causing disrepute to their individual personas.

• With the European Union taking the lead, several governments across the world have taken the initiative to address the concerns arising out of AI statutorily. It has triggered the need for organisations and statutory bodies to adopt policy driven approach towards self-governance. In this regard, the New Zealand government has declared that its current regulatory framework already has adequate guardrails in respect of AI and a standalone AI Act is not necessary. The need for an innovation-oriented policy framework to be “light-touch and risk-based” was emphasised.

Contributors: Joel Bock (US), Jennifer Cass (UK), Sunita Kaur Chima (Malaysia), Robyn Chatwood (Australia), Michael Franzinger (US), Güneş Haksever (New Zealand), Richard Keady (Hong Kong), Dong-Hwan Kim (South Korea), Catherine Lee (Singapore), Julian Ng (Hong Kong), Nadia Ormiston (New Zealand), Davin Olen (South Africa), Andre Rahadian (Indonesia), Constantin Rehaag (Germany), Jenni Rutter (New Zealand), Aliya Seitova (Kazakhstan), Shahid Sulaiman (South Africa), Minh Tran (Vietnam), Linh Tran (Vietnam), David Wagget (UK).