I. Introduction
The global economy is presently experiencing a significant industrial revolution, one propelled not by steam or steel, but by Artificial Intelligence (AI). The swift advancement of technology has created an ecosystem defined by interconnected global supply chains involving data centers, semiconductors, and substantial energy infrastructure[1]. As a result, the companies with the highest valuations today primarily derive their worth from intangible assets, including software, algorithms, data models, and digital content, which are either created or substantially enhanced by AI systems. In response, international arbitration has emerged as a preferred mechanism for resolving disputes. Yet, what makes this shift compelling is that AI is not just creating new disputes, it is exposing the limits of the very architecture designed to resolve them. Arbitration promises a single, efficient forum for global disputes, but it operates within a legal world where intellectual property rights remain strictly territorial and rooted in human creativity. This creates a fundamental tension. This paper reacts to the question: Can international arbitration truly resolve disputes involving AI-generated intellectual property, or is it merely managing a problem it cannot fully solve?
The analysis unfolds in three stages. First, it shows that AI-generated outputs unsettle the territorial foundations of patent and copyright law. Second, it argues that although arbitration is highly effective in resolving the contractual and commercial consequences of AI disputes, it cannot authoritatively settle broader public-law questions of inventorship, ownership, and infringement across jurisdictions. Third, it considers whether developments in SEP/FRAND arbitration offer a limited but useful model for coordinating cross-border AI disputes, even if arbitration cannot fully replace courts or legislators. Ultimately, “This paper argues that unless arbitration evolves from isolated bilateral agreements into a multilateral 'matrix of consent,' it will remain paralyzed by the territorial fictions of traditional IP law."
II. The Structural Limit: Territorial IP Frameworks vs. Autonomous Global AI
A foundational principle of intellectual property law is that rights are strictly territorial, meaning that legal protections, definitions, and enforcement mechanisms vary significantly from one jurisdiction to another. Historically, this fragmented system functioned adequately because human authorship and inventorship provided a universally recognized and stable foundation for IP rights across borders. However, AI fundamentally disrupts this paradigm because it operates globally and seamlessly, challenging the core philosophy of global patent law: the "Incentive and Reward Concept".[2] Since the 1474 Venetian statute, the patent system has existed to incentivize and reward human ingenuity and labor.
This structural mismatch was exposed in the Thaler (DABUS) litigation. Dr. Thaler sought to name his AI system, DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), as the sole inventor of two patentable inventions. Courts across jurisdictions, consistently rejected this premise. The UK Supreme Court, analyzing Section 7(3) of the UK Patents Act 1977, ruled unanimously that an "inventor" must be a natural person, dismissing the notion that a machine could hold or transfer a patent.[3] Dr. Thaler's attempt to use the doctrine of accession arguing he owned the "fruits" of the machine just as a farmer owns a calf produced by a cow was firmly rejected. Similarly, the Full Federal Court of Australia held that the grant of a patent is a reward for the ingenuity of an inventor with a legal personality.[4] For international arbitration, the Thaler consensus reveals a devastating legal fragmentation. If an AI system cannot be legally recognized as an inventor or author, it begs the question: who owns the highly valuable outputs generated by these machines? Different jurisdictions may adopt slightly different approaches to AI-generated works; for example, one jurisdiction may recognize copyright in AI-assisted works where there is sufficient human input, while another may deny protection entirely if the work is not "original" in the traditional sense.
In a cross-border dispute over AI-generated code or products, an arbitral tribunal may be forced to navigate a maze of territorial laws where the same AI output might be protected under one jurisdiction’s specific definitions but denied protection in another. Arbitration seeks to provide a unified resolution but operates strictly within the confines of applicable national laws, it cannot harmonize these differences. At most, it relocates legal uncertainty from the public courtroom into a private forum, while leaving the underlying question of global entitlement unresolved. That unresolved question does not remain abstract for long. In practice, it quickly resurfaces inside licensing agreements, data-sharing arrangements, and technology supply contracts.
III. Contractual Spillover : Tech Supply Chains and Expansion of IP Conflict
The second significant challenge to the effectiveness of arbitration stems from the integration of artificial intelligence within global supply chains and commercial licensing agreements. Contemporary multinational corporations depend on an extensive array of copyright assignments, trademark licenses, and patent agreements. Consequently, many AI-related disputes begin as standard contractual disagreements such as disputes over licensing terms, usage rights, or data-sharing obligations but rapidly metastasize into massive, multi-jurisdictional IP conflicts.[5]
Envision a situation in which a technology firm grants a license for an artificial intelligence software model to a subsidiary located abroad. The software is utilized to generate content or execute functions that depend on extensive, scraped datasets. Subsequently, it becomes apparent that the AI system was developed using copyrighted material without obtaining the necessary authorization from the original creators. What initially presents itself as a straightforward violation of licensing terms swiftly transforms into a multifaceted intellectual property conflict.[6] A real world example is the lawsuit filed by Getty Images against Stability AI, claiming that over 12 million photographs were scraped and copied to train the "Stable Diffusion" AI without proper permission or licensing.[7]
In resolving these disputes, Alternative Dispute Resolution (ADR) mechanisms like arbitration are highly favoured over litigation for several key reasons. First, AI "black box" algorithms require specialized knowledge, and arbitration allows parties to select neutral decision makers with dedicated technical IP expertise, avoiding generalist court judges who may lack technological fluency. Second, because AI cannot secure patent protection in many jurisdictions, developers increasingly rely on trade secrets to protect their training algorithms. Arbitration provides a secure, confidential environment, shielding proprietary source code from the public domain. Despite these vast procedural advantages, arbitration faces a severe limitation: a tribunal's mandate is drawn entirely from the contract's arbitration clause. While a tribunal is perfectly equipped to interpret contractual obligations and resolve the private contractual breach between the licensing parties, it lacks the jurisdictional authority to definitively resolve the underlying, public IP issues that extend far beyond the scope of the private contract. For instance, a tribunal may need to determine whether a party breached a licensing agreement by using AI-generated outputs, without clear guidance on whether those outputs are actually protected by IP law. Since arbitration is forced to leave these core, borderless IP issues to territorially bound public courts, it suffers from a deeper, systemic flaw when adjudicating AI, a profound legitimacy deficit.
IV. A Qualified Response: Can SEP/ FRAND offer a Model for AI Disputes
The systemic challenges delineated above highlight a more extensive critique concerning the limitations of private, confidential tribunals in addressing issues of significant public interest. Arbitration is essentially a private, consent-driven mechanism intended to address specific disputes arising between contracting parties. Nevertheless, the challenges arising from the AI revolution specifically those concerning inventorship, authorship, and the extent of intellectual property protection are fundamentally public in character. However, the perceived legitimacy deficit rooted in confidential arbitrations that undermine public accountability and regulatory transparency, is being tested and potentially mitigated by the arbitration sector’s success in managing Standard Essential patent (SEP) and fair, reasonable and non-discriminatory (FRAND) disputes.
While critics contend that delegation of AI and global technological standards to private dispute resolution is problematic, the realities of contemporary technology chains offer a persuasive counter narrative. The AI industrial revolution is not taking place in isolation, it is rather founded upon interconnected global infrastructure that encompasses semiconductors, extensive data centres and advanced telecommunications network. These foundational technologies are frequently governed by complex global IP licensing programmes and SEP/FRAND frameworks. Pursuing ‘piecemeal litigation’ for such global IP disputes in diverse national courts is exceedingly complex, cumbersome and expensive. Consequently, there has been a pronounced trend toward utilizing international arbitration to resolve multi-jurisdictional FRAND disputes in a single, consolidated forum, thereby avoiding the inefficiencies and risks of contradictory national courts decisions[8].
This established SEP/ FRAND arbitration, framework provides a vital, practical blueprint for the AI era. Standard setting organizations and authorities in the United States and Europe now actively support the use of arbitration to determine FRAND licensing terms, indicating a growing institutional trust in these private forums to manage broad industry standards. Rather than operating completely devoid of public policy considerations, arbitration in the SEP/FRAND context and increasingly in AI disputes leverages highly specialized arbitrators who possess the specific technical and legal expertise required to navigate these complex fields. By adapting the FRAND model, tribunals resolving AI disputes can utilize the flexibility of arbitral procedures to actively account for the broader interests of market participants and the global public. For that reason, SEP/FRAND arbitration should be understood not as a full blueprint for AI governance, but as evidence that arbitration may still serve a pragmatic coordinating function at the edges of a fragmented legal system.[9]
V. Conclusion: The Limits of Private Justice in a Borderless Revolution
The collision of artificial intelligence and intellectual property has fundamentally outgrown the architectural confines of both territorial national laws and private dispute resolution.We are currently attempting to govern a 21st century, borderless, non-human intelligence using a 15th century framework of human centric property rights. Compounding this friction, we rely on a mid-20th century mechanism of private dispute resolution to quietly sweep the resulting fallout under the rug. As this analysis reveals, international arbitration serves as an invaluable pressure valve efficiently managing the commercial fallout of global algorithmic innovation. Yet, they are ultimately applying a private bandage to a public rupture.
The path forward requires abandoning the traditional, isolated arbitration clause in favor of "ecosystem-wide" dispute frameworks. Just as the tech industry relies on open source licenses and complex webs of Standard Essential Patents to function, AI developers, data providers, and commercial end-users must embed multilateral arbitration agreements directly into the foundational architectures of their AI models and application programming interfaces (APIs). By weaving an interconnected, mandatory matrix of consent throughout the entire digital supply chain, arbitration can transcend its isolated inter partes limitations and approximate the erga omnes reach of public courts. Furthermore, arbitral institutions must pioneer specialized, transparent "Tech Appellate Boards" that publish redacted jurisprudence, actively cultivating a new lex mercatoria (commercial law) for artificial intelligence that public regulators are currently too slow to write. We can no longer afford to view international arbitration merely as a confidential escape hatch from contradictory national laws. This raises a larger question, that remains unsettled.
As artificial intelligence autonomously reshapes the global economy and generates unprecedented intangible wealth, will the international legal community possess the vision to transform arbitration into a borderless, multilateral architecture of justice, or will we remain paralyzed, clinging to the territorial fictions of the past while the algorithms we created dictate the rules of the future?
References
[1]Patrick Schroeder et al., AI’s Industrial Revolution: A New Frontier for Disputes, Freshfields (Feb. 2026)
[2] Tan Tee Jim SC, Artificial Intelligence as Inventor?, 36 Sing. Acad. L.J. 346 (2024).
[3] Pramesh Prabakaran, Man versus Machine: UK Supreme Court Upholds Requirement for Human Inventorship to Secure Patent Protection, NUS L. (June 2024).
[4] Thaler v. Comptroller-Gen. of Patents, Designs & Trade Marks, Bus. L.R. 47 (UK)
[5] Hans-Patrick Schroeder et al., AI’s Industrial Revolution: A New Frontier for Disputes, Freshfields (2026).
[6] David J. Kappos, Sharonmoyee Goswami, Teh Joo Lin & Adriana Uson, Making the Case for Arbitration in Global AI IP Disputes, IAM (Dec. 18, 2024).
[7] Getty Images (US) Inc v Stability AI Inc Case 1:23-cv-00135-UNA (US District Court for the District of Delaware, 2023).
[8] John V. H. Pierce & Pierre-Yves Gunter eds., The Guide to IP Arbitration (2d ed., Law Bus. Res. Ltd. 2022).
[9]Id.