Authors
Introduction
As is evident from other sections of this guide, legal issues arising from AI are wide-ranging and complex, impacting numerous different areas of law. Beyond issues that relate to specific intellectual property rights, the proliferation of AI systems and their potential uses – ranging from medical diagnosis to crime detection – only exacerbate the complexities surrounding AI regulation.
Despite developing case law in a number of jurisdictions, neither legislation nor judicial decisions have, at present, definitively resolved all of the pending issues. This, in combination with some arguably conflicting decisions, leaves a host of uncertainties and unanswered questions surrounding the use of AI systems and AI-generated content. Given that the approach to AI law differs across jurisdictions, we can also expect more jurisdictional issues to arise in AI disputes. We address some of the issues that have arisen, and others that are likely to emerge, below.
Determining liability
Once a rights holder has identified a potential infringement of their rights or other unlawful conduct for which compensation may be payable, one of the primary challenges will be to identify the party who is legally responsible when AI is involved. This can be a complex issue, as liability may lie with different parties involved in the AI’s development, deployment, or operation, such as the creators, operators, users, or owners of the system. Additionally, AI is often the result of collaboration between multiple parties, such as the AI developer, user, owner, and/or manufacturer. Determining who is responsible for any potentially unlawful acts can be challenging and requires a thorough understanding of the roles and responsibilities of each party involved.
This is a particularly pertinent issue given the recent attempts by some to move closer toward legal personality for AI systems, such as AI pioneer Dr. Stephen Thaler’s unsuccessful multi-jurisdictional litigation aiming to convince courts to recognize the ability of an AI system to be named as the inventor of a patent, or his (so far unsuccessful) attempt to have an AI system recognized as the author of a copyright work in the United States. On the other side are campaigns such as the Human Artistry Campaign, which aims to ensure that copyright protects only human intellectual creativity.
For some causes of action, the state of mind and intention of the defendant is relevant to establishing liability, or to the calculation of damages to which the claimant is entitled. Examples of this include malicious falsehood (sometimes referred to as trade libel) and the calculation of aggravated damages in defamation cases. These concepts may have to be interpreted in a manner that allows them to remain applicable to reputational disputes involving AI.
There have been numerous AI defamation claims brought based on AI outputs, ranging from cases relating to AI systems confusing two individuals with the same name, thereby causing damage to reputation, to hallucinations suggesting that a company had been sued for deceptive practices, leading to reputational harm and lost business. Claims often raise unresolved questions about the application of traditional defamation principles to AI-generated falsehoods across territories, requiring case-by-case assessments. Many are still making their way through the courts.
The EU AI Act (Regulation (EU) 2024/1689) is a comprehensive legal framework for AI, and adopts a risk-based approach to regulate AI safety, transparency, and fundamental rights across the European Union single market. It creates obligations for providers and deployers of AI systems that are dependent on risk profile. High-risk AI systems (including those used in critical infrastructure, employment, and law enforcement) are subject to stringent requirements regarding transparency, data governance, human oversight, and record-keeping, with high financial penalties for non-compliance.
There is currently no comprehensive federal AI legislation in the United States dealing with liability, and most legislative activity has been at state level. Reports indicate that more than 1,500 AI-related bills have been introduced across all 50 states on topics ranging from explicit deepfakes to accountability for output, and some have already been enacted at state level. Still, some of the big questions relating to liability and training remain largely unanswered, although cases such as Bartz v. Anthropic and Kadrey v. Meta do provide some guidance, particularly on fair use (training on lawfully acquired materials is more likely to be lawful, evidence is vital, and the effect on the market looks likely to be a significant consideration).
In the UK, the white paper on AI regulation (published in March 2023) proposed a pro-innovation, proportionate, trustworthy, adaptable, clear, and collaborative AI regulation regime. However, as of 2026, the UK still does not have any AI-specific regulation or legislation covering AI as a technology. The Intellectual Property Office’s Report on Copyright and Artificial Intelligence, following a consultation, adopted a wait-and-see approach – there are no longer plans to introduce a broad text and data mining exception. Instead, the courts and AI subject-specific legislation will be left to determine matters for the time being.
Establishing jurisdiction
Another challenge in enforcing rights infringed by an AI system is determining the jurisdiction in which legal action can be commenced. Absent any clear, universal legal framework in this regard, this must be assessed on a case-by-case basis, which is likely to lead to forum shopping by claimants. Where an AI system is developed and/or deployed in different jurisdictions, it may be challenging to determine which jurisdiction’s laws and regulations apply.
While certain actions can be commenced in the territory where the damage is suffered (under the so-called “accessibility criterion” in the EU, for instance), others require determining where the infringing acts were committed or where the defendant is situated. This can prove challenging for disputes involving AI systems, as it is not always easy to establish, for example, in which territory the relevant acts are deemed to have been carried out.
In the EU, there were proposals for common rules for Member States under a non-contractual civil liability regime relating to damage caused by AI, particularly high-risk AI systems. However, these were withdrawn in 2025 due to lack of stakeholder agreement.
The importance of jurisdiction in AI disputes is illustrated by the UK court’s first instance decision in Getty Images v. Stability AI [2025] EWHC 2863 (Ch), the first major UK judgment addressing copyright and trademark issues in the context of AI model development. The judgment deals predominantly with the claimant’s allegation of secondary copyright infringement (which was rejected) and historic trademark infringement relating to watermark reproduction (which the court agreed had taken place). Significantly, Getty had to abandon its copyright infringement claims based on training because the relevant model training had occurred outside the UK. Thus, territoriality is a matter of critical importance in AI disputes, at least in the UK, for the time being.
Recent developments in Germany in the GEMA v. Suno case suggest that the German courts may be willing to accept jurisdiction over training carried out in the United States even without expert evidence on U.S. law – the court differentiated this case from the U.S. fair use cases on the basis that here, the outputs were substantially similar to the input content and were generated from non-specific prompts.
AI and evidentiary considerations
Reliance on AI-generated evidence
It may be tempting in some situations to use AI-generated evidence to support a party’s position in court. This can, however, give rise to challenges related to the admissibility of such evidence. For example, if the AI system uses proprietary algorithms or data that are not disclosed or accessible to the other party, it may be difficult – or even impossible – to challenge or verify the results produced by AI.
Even if it is admissible, AI-generated evidence may not yet be considered credible or convincing. AI is often said to be a “black box,” meaning that it can be difficult to understand how it reached a particular decision or outcome. Unlike a human witness, AI systems cannot (yet) be reliably cross-examined, so one or more expert witnesses may be required to enable the court to understand how the relevant AI system works and how it arrived at a given decision. This can be costly and time-consuming, particularly given the expertise, experience, and knowledge required to testify on such matters. It is also unclear who would bear the burden of proof in such instances, that is, whether it would be for the party that intends to rely on AI-generated evidence to provide sufficient information about the relevant AI system or whether it would be for the party challenging this evidence to introduce doubt.
Ethical and data protection concerns related to AI must also be taken into consideration when relying on AI-generated evidence. It may be necessary to present additional evidence and arguments to demonstrate that the use of AI was ethically justifiable and compliant with applicable data protection obligations.
Until such time as the admissibility and weight afforded to AI-generated evidence have been addressed by the courts or in statute, it is difficult to predict how these issues will be resolved.
Discovery, disclosure, and privilege
Similar considerations and risks will apply to the process of discovery or disclosure (as they are known in the United States and the UK, respectively) and to formalized evidence-gathering exercises in other territories. AI-driven automated processes are likely to play an increasing role in these phases of litigation, so it may be that certain rules and principles – such as those governing agreed search terms and searching methodologies – will need to be reconsidered to address the challenges and opportunities of AI.
This process is already underway in the UK. The Civil Justice Council working group in England and Wales (chaired by Lord Justice Birss, one of the first English judges to openly use AI to assist with drafting one of his judgments) has published its interim report and consultation titled The Use of AI for Preparing Court Documents. It examines the implications of AI for various procedural stages of court litigation, considers whether AI-specific rules and guidance should be put in place, and suggests that declarations by legal representatives may be required relating to the use of AI in specific circumstances, for example in generating evidence using AI.
The question of privilege in documents generated by AI has been considered in the U.S. cases of Warner v. Gilbarco, Inc. and United States v. Heppner, which reached different conclusions as to whether those documents would be privileged. In the UK case of Munir v. Secretary of State for the Home Department, the Upper Tribunal indicated that uploading documents into public AI tools both breached client confidentiality and waived privilege.
Broader risks to the credibility of evidence posed by AI
AI-generated content, including sound-alikes and deepfake videos, is becoming sufficiently elaborate to credibly and successfully pass for authentic or human-generated content. This poses a significant threat to the credibility of evidence generally, as parties may seek to discredit adverse evidence by claiming that it is – or could be – AI-generated content. This is no longer a merely hypothetical scenario: in the Californian case Mendones v. Cushman & Wakefield, Inc., the plaintiffs’ claim was dismissed once it became apparent that they had submitted AI-generated videos of a woman providing oral evidence.
There has, however, been a significant increase in the legislative response to the deepfake threat. In April 2025, U.S. Congress passed the TAKE IT DOWN Act, a bill that criminalizes the nonconsensual publication of intimate images, including “digital forgeries” (i.e., deepfakes) intended to cause harm, in certain circumstances, and requires platforms to implement a “notice-and-removal” process to remove such images at the depicted individual’s request. Similar legislative action has been taken in the UK through the combined operation of the Online Safety Act 2023 and the Crime and Policing Act 2026, which together create criminal offenses and platform takedown obligations materially comparable to the TAKE IT DOWN Act.
The EU AI Act includes transparency obligations requiring that AI-generated or manipulated content (including deepfakes) be clearly labeled, and it may be that we will start to see similar rules in other jurisdictions, particularly relating to evidence (see the Civil Justice Council’s working group in the UK).