This contribution focuses on the application of artificial intelligence (AI) in cross-border project finance transactions, examining both current uses and the legal challenges they present with potential private international law implications.

Firstly, it should be noted that project finance has been relatively slow to adopt AI in an integrated fashion. This may be due to structural characteristics of this type of transaction, difficulties in collecting sufficient, reliable, information from particular industries and regions, and the fact that current technologies are not yet fully capable of addressing the complex legal nuances often involved in project finance that still require human judgment and subjective interpretation (for example, the use of reasonableness and good faith qualifiers in contractual language and its legal implications or human factors influencing political risks). The result is that project finance largely continues to rely on manually drafted and reviewed voluminous documentation, complex spreadsheets, and analogous calculations.

The above said, certain applications of AI are already present in the different stages of a project finance, even if on a siloed basis, in particular in the renewables sector. This presence is expected to grow exponentially in the coming years as technology rapidly evolves and more solutions are developed to suit the unique characteristics of international project finance transactions in different regions and sectors.

In the area of risk assessment and feasibility analysis, for example, AI tools are increasingly used to process public and proprietary data to enhance risk identification and support feasibility studies. In energy projects, tools are applied to estimate power output and demand variations—even beyond what humans are capable of detecting—to support enhanced modeling, market trend assessments, and thereby inform financial analysis.

In the due diligence context, document review can be significantly streamlined by AI tools (in particular, traditional tools focused on data processing and summarization, rather than data generation). The use of AI can free human resources to focus on areas where subjective analysis is required. AI can also be used to track regulatory compliance obligations, identify red flags (and automatically notify stakeholders), and monitor human rights and financial misconduct risks as projects are implemented.

AI tools can similarly be used to assist with contract management. This is increasingly the case in the construction industry where AI is being applied to track contractual milestones, also at times through integration with smart contract systems. Some platforms can also evaluate external data—such as real-time weather or geopolitical indicators—to assess the probability of force majeure events or other disruptions.

In the field of environmental, social and governance (ESG), AI-based systems can also be used to monitor metrics and support reporting and required disclosures.

One noteworthy tool is MALENA, developed by the International Finance Corporation (IFC). Drawing on the IFC’s extensive project database, MALENA allows users to upload project documentation and benchmark it against historical data to flag reputational risks, monitor emissions, and support ESG disclosures. The tool is publicly accessible and has applications for both investors and project sponsors.

Finally, the most technologically advanced applications of AI are emerging in the operational phase. For example, the development of “smart grids” where AI-enabled systems are capable of dynamically optimizing energy distribution. This enhances efficiency, reduces energy waste, and ultimately bolsters grid reliability.

Another impressive example is the US Department of Energy’s Argonne National Laboratory’s development of an AI solution capable of improving the design and operational performance of nuclear reactors. By processing historical and real-time performance data, these systems can identify early warning signs, refine maintenance schedules, and lower operational risks and costs.

The list goes on with each passing day.

The potential of AI applications is therefore clear: these tools can support more accurate risk analysis, save time and resources, and reduce the need for certain third-party services. They can assist in managing regulatory compliance across multiple jurisdictions—a particularly complex challenge in project finance—and facilitate audits and diagnostics when problems arise. Ultimately, AI can contribute to optimizing both project development and long-term operations.

However, in addition to its evident benefits, the use of AI presents unique legal questions and challenges, particularly when it comes to assessing and attributing liability in the event of harm derived from its deployment and use.

One key issue is attribution. AI tools typically involve multiple parties in their development, data collection, technology deployment and use. If harm occurs, it therefore can be complex to attribute legal responsibility among these parties, especially where roles are fragmented and data provenance is unclear.

A closely related challenge is that of jurisdiction. In cross-border transactions involving parties in multiple jurisdictions, it becomes difficult to determine who is liable if harm occurs in a location distinct from where any of the actors are based. Determining applicable law and competent courts is made even more complex by the fact that legal frameworks for AI remain under development and vary significantly across jurisdictions.

Another relevant legal issue is causation. A defining characteristic of AI—especially machine learning models—is what is often referred to as the “black box” problem. While these systems may display a form of intelligence, it should be noted that it is not human intelligence. This means that even the technology creators may not fully understand the logic or reasoning behind their outputs. This opacity can make it difficult to determine at what point harm occurred: during design, data input, system deployment, or possibly as a result of external interference such as a cyberattack. These questions are further complicated when humans are involved in reviewing or implementing AI outputs, potentially breaking the chain of causation.

Along the same lines, cybersecurity is separately an area of concern. AI systems increase the attack surface, and when used in critical infrastructure—as in the Argonne Laboratory example—the consequences of a cyberattack could be severe. Moreover, systemic failure could pose risks to public safety, and the legal implications would be wide and extensive.

Accordingly, evolving legal frameworks must anticipate such scenarios and establish adequate risk allocation mechanisms.

This includes both the continued development of applicable regulations and the evolving interpretation of those regulations by enforcement authorities and courts, as well as the application of existing legal principles to the context of AI by analogy. With respect to the latter, the regulatory landscape is expected to remain unsettled for the foreseeable future—particularly in cross-border transactions—making the way enforcement bodies and courts approach AI a critical area to watch.

In the meantime, and given that both project finance and cross-border transactions are fundamentally contractual in nature, many of these risks can and should be addressed through transaction documentation. At present, this remains the most effective tool for managing AI-related risks in project finance. Key recommendations include:

1. (1) Disclosure. Parties should first and foremost disclose AI use, especially where sensitive or critical systems are concerned.

2. (2) Testing and validation. Additionally, contracts may include obligations to test and validate AI tools and processes applied.

3. (3) Auditability. Documentation and audit trails should be maintained to facilitate accountability and traceability.

4. (4) Warranties and indemnities. Transaction documents may allocate risk through appropriate warranties and indemnities regarding functionality, reliability, and the ethical use of AI tools, including data integrity and bias mitigation.

5. (5) Certification mechanisms. Contracting parties may consider requiring that certain AI systems used in projects obtain recognized third-party certifications to verify functionality, safety, or compliance with applicable standards. Certification mechanisms, where available, can offer a valuable additional layer of assurance and accountability even where not mandated by applicable law.

These measures can assist in clarifying responsibilities, allocating risk, and enhancing accountability in the use of AI within project finance.

This contribution examines five hypothetical scenarios to analyze the U.S. legal framework for establishing jurisdiction over foreign defendants in copyright infringement cases, with a focus on identifying legal principles relevant to the potential liability of foreign artificial intelligence (AI) companies under U.S. copyright infringement law.

In practice, approximately half of all copyright lawsuits against foreign defendants in U.S. courts are dismissed for lack of jurisdiction. Against this background, the following hypothetical explores when a foreign AI company might be subject to jurisdiction in U.S. courts—particularly in the context of alleged copyright infringement.

Notably, while recent domestic lawsuits, such as New York Times v. OpenAI, exist, there are few published decisions available. Accordingly, the following analysis draws on comparable cases to identify relevant principles of law that can be used by courts by analogy, which is a key element of common law.

Under U.S. law, copyright is governed by federal statute. Federal statutes are presumed not to apply extraterritorially absent clear language to the contrary, which the Copyright Act does not contain. Therefore, for a foreign defendant in a copyright dispute to be subject to jurisdiction in a U.S. federal court, the plaintiff must establish a connection to the United States at the domestic level. This is referred to as a predicate act. In simple terms, the plaintiff must establish an act—a predicate act—of copyright infringement that occurred domestically and connects to the alleged extraterritorial conduct. This principle applies equally in the context of trademarks and other IP infringement disputes.

Establishing this domestic predicate act requires an analysis of personal jurisdiction, specifically whether the foreign defendant has “purposefully availed” itself of the U.S. market. We therefore consider five illustrative scenarios reflecting varying degrees of contact with the United States.

This contribution builds on the prior discussion of jurisdictional questions in copyright matters, including references to BBC, TV Polska, and other relevant cases.Footnote ¹ It discusses private international law issues at the intersection of AI and copyright law.

Three categories of legal issues typically arise at this intersection. The first concerns whether works generated by AI are protectable under copyright law. The second relates to whether the unauthorized use of copyrighted works for training AI models may give rise to legal liability. The third, which receives comparatively less attention, involves the use of AI for enforcement, commercialization, licensing, and other related purposes. To illustrate these issues, this contribution presents two scenarios.

This contribution explores the historical foundations of private international law and their relevance to current transformations driven by emerging technologies within the context of state authority and legitimacy. Rather than focusing solely on artificial intelligence (AI), it therefore suggests situating AI within a broader category: general-purpose technologies.

The decline of the nation state is a familiar observation in contemporary legal and political discourse. The key question is whether we are now facing a fundamental transformation—qualitative rather than quantitative—in the nature of state power. If so, what would serve as the source of legitimacy in the emerging global order?

The argument advanced here is that AI is best understood as one catalyst within a broader technological transformation.

It is worth recalling that the roots of private international law may be traced back to the jus gentium of Roman law—a legal order that governed interactions among states within an empire, but without modern nation states. The Restatement (Third) of the Foreign Relations Law of the United States defines private international law as “State to State,” reinforcing its embeddedness in the nation-state system—a characteristic of modern international law more broadly.

The economic concept of “general-purpose technology” (GPT) refers to technologies with wide-reaching and long-term transformative effects, such as fire, agriculture, writing, the printing press, and the steam engine. A list of technologies that proved to qualify as GPT over time, as identified by economists, is readily available online. Once we recognize that such technologies tend to catalyze broader social changes over long periods of time, it becomes popular to try to identify the next one. As a result, many things have been prematurely labeled as the next general-purpose technology, only to fall short.

American researcher, Roy Amara, famously observed that people tend to overestimate the short-term impact of new technologies and underestimate their long-term consequences. Gartner’s Chief of Research, Chris Howard recently popularized a version of this idea in the Gartner Hype Cycle, which illustrates the trajectory from overenthusiasm to quiet assimilation. Howard noted that technologies often have their greatest impact once they become invisible. That is, when everyone is talking about a new technology, it often has no real impact. It is only when that technology is assimilated into institutions—and is no longer the subject of discussion– that it begins to have profound effects. Practical examples include containerization and electronic data interchange in global supply chains, both of which have significantly shaped globalization while remaining largely unrecognized by the general public.

It should be noted that Gartner is a consulting firm that helps large corporations make responsible technology acquisitions and is known for coining memorable terms. In 2012, Gartner coined the term “nexus of forces” to describe the convergence of cloud computing, big data analytics, social production, and AI. These technologies—though distinct—act collectively as engines of transformation. For example, a mobile phone today vastly outperforms the mainframe computers of the 1970s and social production now necessarily encompasses social media and collaborative platforms, such as Uber and Lyft. Economists who previously identified factories as general-purpose technologies were referring to similarly complex systems.

Bill Gates, in his 1995 book The Road Ahead, articulated a basic rule: automation applied to an efficient process enhances efficiency; applied to an inefficient one, it compounds dysfunction. This observation remains relevant for legal institutions. The legal profession’s adoption of technology in the United States has, in many cases, exacerbated access-to-justice issues rather than alleviated them. Courts across the country now routinely process matters in which one or more parties are unrepresented due to cost barriers. Applying AI to such systems without structural reform risks worsening these disparities.

Charles Tilly, a sociologist, once characterized the nation state as an organized crime protection racket—the use of the state apparatus by a small group for resource extraction under the guise of legitimacy. He referred to this concept as “state capacity.”

For example, the concept may explain Donald Trump’s election. Many Americans perceived the Democratic Party as running an organized crime operation and therefore believed they were entitled to have their own representatives in power. In other words, the rise of populist movements can be understood in part as reactions to perceived elite capture, where opposing groups seek control over state mechanisms based on a similar rationale. To counter this, it is suggested that we should stop mythologizing the nation state.

In 2008, Charles Sabel and Jonathan Zeitlin spoke of a “Cambrian explosion” of private ordering. Compared to the mid-twentieth century, we are now witnessing an exponential proliferation of alternative governance structures, many operating outside traditional state frameworks.

Herbert Stein, former chairman of the Council of Economic Advisers, famously remarked: “If something cannot go on forever, it will stop.” It is posited that the real question is not whether but when. How many panels can one attend on the decline of the nation state and the rise of non-state ordering before we acknowledge that we are witnessing a categorically different social order?

The question then becomes: how might legitimacy be constructed in a post-nation state context? Modern legal and political systems rest on the legitimacy of the nation state, the market and the rule of law. The English legal system is sui generis in this regard. Before the English Civil War, Sir Edward Coke, then a judge on the Court of Common Pleas, articulated a theory—widely shared at the time—that customary law had formed the foundation of English social order since time immemorial. This ideology formed part of the constitutional monarchy’s foundation. Though Coke’s account is historically inaccurate, it illustrates the ideological construct underpinning that era.

The “nexus of forces” is accelerating institutional change. As the decline of the nation state accelerates, the battleground may increasingly concern the definition of knowledge itself. This raises questions of epistemic culture and knowledge governance. Modern law and public international law both presuppose the existence of a nation state. As we undergo a knowledge governance revolution, how do we distinguish legitimate from illegitimate power?

There is a compelling discussion of this topic in Garry Kasparov’s book Deep Thinking. He argues that knowledge workers today fear AI and similar technologies because of their disintermediating effects. Drawing a parallel with the nineteenth-century elite reaction to the displacement of labor during the Industrial Revolution, Kasparov writes that either we acknowledge the errors of the past, or we stop complaining in the twenty-first century.

If modernity had a beginning, as Herbert Stein suggested, it will also have an end. What succeeds modernity is unlikely to be “post-modernity,” just as the Renaissance was not merely “post-feudalism.” Naming this new era is still an open question—but an essential one.

In this context, the central question for private international law is what role it might play in the governance of knowledge and institutional legitimacy beyond the nation-state model. One possibility is that it may evolve toward a system more akin to customary law—as understood by Coke and his contemporaries—though now transposed into global, digital, and non-state forms.

More importantly, what will replace modernity? One possibility is that global technology platforms will take on roles once held by states. In this scenario, the platforms and the nation states resemble antagonists, vying for dominance, where voluntary, decentralized standard-setting bodies and certification mechanisms may represent early examples of twenty first-century customary law.

Global technology platforms—operating above or alongside states—are reshaping how norms are developed and enforced. The Chinese Communist Party, operating above the nation state, may be the most successful large-scale example of such a system. Further examples may yet emerge.