THE AUTHOR<\/strong>: The introduction of artificial intelligence (\u201cAI\u201d<\/strong>) into the world of investment treaty arbitration (See e.g. here<\/a>) raises novel jurisdictional challenges for investor-state dispute settlement (\u201cISDS\u201d<\/strong>) tribunals. The present article focuses on issues that may arise in a tribunal\u2019s analysis of its subject-matter jurisdiction in an AI-related investment treaty case.<\/p>\n\n\n\n This article will first briefly set out the general legal framework applicable to the determination of subject-matter jurisdiction in ISDS proceedings (1)<\/strong>. Thereafter, it will analyse the conditions that an AI-related economic activity must satisfy to fall within an ISDS tribunal\u2019s subject-matter jurisdiction (2)<\/strong>, (3)<\/strong>, and (4)<\/strong>. <\/p>\n\n\n\n As reflected in ISDS jurisprudence, an investment treaty tribunal must, in order to ascertain jurisdiction ratione materiae<\/em>, verify the satisfaction of at least three cumulative conditions:<\/p>\n\n\n\n The following section examines the challenges that tribunals may encounter in determining their subject-matter jurisdiction in AI-related ISDS proceedings when addressing each of these ratione materiae<\/em> questions.<\/p>\n\n\n\n To verify the existence of rights or interests in property constituting an \u201cinvestment\u201d, an ISDS tribunal must assess whether such alleged rights or interests are recognised by the laws of the host state (See Zachary Douglas<\/a>, The International Law of Investment Claims52-72; and Reza Eftekhar<\/a>, The Role of the Domestic Law of the Host State in Determining the Jurisdiction ratione materiae<\/em> of Investment Treaty Tribunals 233-253).\u00a0<\/p>\n\n\n\n Generally speaking, municipal property laws have not kept pace with developments and innovations in modern intangible and digital asset forms. As such, in certain cases, it may be difficult for innovative forms of assets to be recognised by the domestic law of the recipient state as \u201cproperty\u201d underlying an investment. In the AI context, while assets such as physical equipment used in data centres may conveniently fall within recognised forms of tangible assets in various legal systems, it is less clear whether \u201cdata\u201d and \u201calgorithms\u201d, as core assets of an AI investor (See Mark McLaughlin<\/a>, Regulating Artificial Intelligence in International Investment Law, The Journal of World Investment & Trade, Vol. 24, Issue 2, 260), would qualify as \u201cproperty\u201d under the municipal laws of the host state at stake.<\/p>\n\n\n\n As to \u201cdata\u201d (in a non-personal context), a great majority of legal systems do not recognise data itself as an autonomous category of property that can be owned. However, through contractual rights, intellectual property (\u201cIP\u201d<\/strong>)-related trade secrets, and database laws, some legal systems recognise certain rights, short of ownership, for the person exercising control over datasets (Mark McLaughlin<\/a>, Regulating Artificial Intelligence in International Investment Law, The Journal of World Investment & Trade, Vol. 24, Issue 2, 274-275). In particular, where the relevant legal criteria are met (e.g., having economic value, secrecy, and adequate safeguard against disclosure), specific datasets and data compilations may fall within existing proprietary regimes (e.g., \u201ctrade secrets\u201d) (See, Article 122 of Iran\u2019s Act on the Protection of Industrial Property<\/a> (2024)). As such, training datasets, as a category of potential asset of an AI enterprise, may\u2014where these conditions are satisfied\u2014constitute a recognised form of asset by the laws of the host state, and thereby capable of forming the basis of a protected investment.<\/p>\n\n\n\n The ISDS tribunal\u2019s award in the pending case of Einarsson v. Canada<\/a><\/em> is poised to address some of these issues. The alleged investment in that case does not concern AI-related data but marine seismic data generated and used in the exploration of oil and gas reserves in Canada. The claimants argue that Canada disclosed their proprietary marine seismic data to third parties and thereby committed, inter alia<\/em>, an uncompensated indirect expropriation (See here<\/a>).<\/p>\n\n\n\n One threshold question that the tribunal will have to address in that case is whether the seismic dataset at issue is a recognised form of property under the municipal laws of the host state. What might pave the way for the Einarsson<\/em> tribunal in this respect is that, in the course of litigation before the Canadian courts, the Alberta Court of Appeal<\/a> found that the claimants\u2019 seismic data was copyright-protected, which indicates that judicial organs of the respondent itself acknowledged that the data in question attracted proprietary protection under domestic law. While not binding on the tribunal (Amco<\/em> v. Indonesia<\/em><\/a> [177]), as each investment treaty tribunal is the judge of its own jurisdiction (Empresas Lucchetti<\/em> v. Peru<\/em><\/a> [81]-[87]), if the relevant conditions are satisfied, an arbitral tribunal may defer to the findings of a municipal court addressing municipal law issues that bear on the tribunal\u2019s jurisdiction (Soufraki v. UAE<\/em><\/a> [96]-[97]).<\/p>\n\n\n\n As to \u201calgorithms\u201d, which likewise do not enjoy explicit recognition in the overwhelming majority of domestic property law regimes, the more likely scenario is that arbitrating investors would characterise them as subject to intellectual property rights protected by host state law. An AI investor may frame its algorithms as copyrights, patents, trade secrets, or even know-how (Mark McLaughlin<\/a>, Regulating Artificial Intelligence in International Investment Law, The Journal of World Investment & Trade, Vol. 24, Issue 2, 271). In each case, the investor must demonstrate that the conditions for the recognition of such intellectual property rights are met under the local law of the receiving state. Depending on the domestic legal system concerned, respondent states may seek to contest the existence of the required conditions. For instance, if an AI investor characterises its algorithm as a \u201cpatent\u201d, the host state may argue that the algorithm in question does not meet an essential condition of patentability under its national law (specifically, the requirement of technical character) (Guidelines for Examination in the European Patent Office (April 2026), Part G \u2013 Chapter II-5 [3.3.1]<\/a>).<\/p>\n\n\n\n All in all, in the absence of explicit recognition under host state law, an ISDS tribunal faces a particular challenge. An arbitrating investor may try to frame its assets as subject to intellectual property rights or otherwise emphasise the economic value, income-generating capacity, and transferability of AI datasets and algorithms. The respondent state, by contrast, may contend that such items lack formal recognition as property under its domestic law and dispute whether municipal law requirements for intellectual property protection are satisfied.<\/p>\n\n\n\n In order to pass this jurisdictional threshold, an ISDS tribunal has to verify whether the assets and the economic operation in question satisfy the subjective and the objective test for an \u201cinvestment\u201d (See Rudolf Dolzer et al<\/em><\/a>, Principles of International Investment Law (3rd Edition) 84). <\/p>\n\n\n\n To satisfy the subjective test, the tribunal must assess whether the assets and the economic operation concerned are covered by the purported definition of the term \u201cinvestment\u201d in the relevant investment treaty (See Rudolf Dolzer et al<\/em><\/a>, Principles of International Investment Law (3rd Edition) 84).<\/p>\n\n\n\n The vast majority of investment treaties contain an \u201casset-based\u201d definition of an investment, typically including an illustrative list of assets, encompassing certain intellectual property rights. With respect to AI assets, whereas more traditional forms of investment, such as shares and stocks in an AI enterprise and physical assets, such as server infrastructure, may face little difficulty being covered by the definition of the term \u201cinvestment\u201d in an investment treaty, more modern features of AI technology, e.g., AI algorithms and AI datasets, may face challenges in this respect. Although investment treaties have undergone significant reforms over the past two decades, chiefly in response to developments in ISDS, provisions defining the term \u201cinvestment\u201d have not been substantially updated to keep pace with technological developments in digital asset forms (Niels Lachmann<\/a>, Opening the Black Box of Data\u2019s Relationship with International Investment Law, ICSID Review – Foreign Investment Law Journal, 2026, 3).<\/p>\n\n\n\n In view of the foregoing, fitting items such as datasets and algorithms within an asset-based definition of \u201cinvestment\u201d incorporated in the underlying treaty might be challenging, in particular, given that \u201cdatasets\u201d and \u201calgorithms\u201d are not usually mentioned by name in the paragraph regarding intellectual property rights. In addition, where the intellectual property rights paragraph is not sufficiently inclusive, it would become even more burdensome to include such assets within the treaty definition of \u201cinvestment\u201d.<\/p>\n\n\n\n Where dubious, therefore, Counsel pleading on behalf of the investor might try to frame assets such as datasets and algorithms as \u201ctrade secrets\u201d (e.g., Article 1(1)(f) of the Georgia-Hungary Bilateral Investment Treaty<\/a> (\u201cBIT<\/strong>\u201d) (2024)) or \u201cconfidential business information\u201d (e.g., Article 1(1)(a)(iv) of the Argentina-US BIT (1991)<\/a>). Under certain modern investment treaties, one may rely upon the \u201cundisclosed information\u201d item incorporated in the \u201cintellectual property rights\u201d paragraph of an investment definition provision (e.g., Article 1(a)(vii) of the Japan-Zambia BIT (2025)<\/a>). Some recent investment treaties include \u201ctechnology\u201d (e.g., Article 1(1)(d) of the Russia-Congo BIT (2025)<\/a>) or \u201ctechnical processes\u201d (e.g., Article 1(1)(d) of Germany-Pakistan BIT (2009)<\/a>) within the paragraph pertaining to intellectual property rights. Investors may avail themselves of the latter two terms to qualify \u201calgorithms\u201d as investments.<\/p>\n\n\n\n Another possibility is to draw on the open-ended nature of the \u201cinvestment\u201d definition in the treaty at stake, utilising the \u201cevery kind of asset\u201d language. This exercise, however, would likely not go unchallenged by the opposing side, who may raise other ratione materiae<\/em> objections (see below). In addition, such an argument would not be feasible in the case of treaties with a closed-list approach to the definition of \u201cinvestment\u201d (e.g., Article 13(1) of Canada – Indonesia \u00a0Comprehensive Economic Partnership Agreement<\/a> (\u201cCEPA\u201d<\/strong>) (2025)).<\/p>\n\n\n\n To satisfy the objective test of an investment, a tribunal should verify whether the specific economic operation in question satisfies the characteristics of an investment (See Rudolf Dolzer et al<\/em><\/a>, Principles of International Investment Law (3rd Edition) 84). These features typically comprise the commitment of capital or other resources, a certain duration, the expectation of gains or profits, the assumption of risk, and the contribution to the economy of the host state. These criteria are commonly referred to as the Salini<\/em> test (See here<\/a> and here<\/a>).<\/p>\n\n\n\n It has been stated time and again that in order to verify the conditions of the Salini<\/em> test, a tribunal has to adopt a holistic approach to the entire economic operation (e.g., Uni\u00f3n Fenosa Gas v. Egypt<\/a><\/em> [6.66]-[6.68]). Thus, in the context of an AI-related investment treaty dispute, it would be incorrect to limit the analysis to the individual components of an AI investment. Rather, to satisfy the objective test of an investment, the tribunal should consider the economic operation as a whole.<\/p>\n\n\n\n In cases where the foreign investment consists of the construction of an AI data centre, the economic operation would be akin to a traditional construction and infrastructure project, typically displaying the hallmarks of an investment under both a subjective (e.g., Article 9(1)(c) of the Comprehensive and Progressive Agreement for Trans-Pacific Partnership (\u201cCPTPP\u201d<\/strong>) (2018)<\/a>, definition of an \u201cInvestment Agreement\u201d) and an objective (e.g., SCB v. Tanzania<\/a><\/em> [213]) test.<\/p>\n\n\n\n Similarly, long-term AI service contracts for the automation of public services in the host country that entail operational and sovereign interference risks may be characterised as investment service contracts, falling within treaty definition (e.g., Article 9(1)(b) of CPTPP (2018)<\/a>, definition of an \u201cInvestment Agreement\u201d) and satisfying the objective characteristics of an investment (e.g., Bosca v. Lithuania<\/a><\/em> [168]).<\/p>\n\n\n\n However, it would not, on all occasions, be straightforward for AI-related activities to fulfil the Salini<\/em> test requirements. In addition, some such investments may find it more difficult to satisfy the territorial nexus requirement. This is particularly so in cases where the AI activity is conducted from abroad rather than within the territory of the host state. In such situations, the state may highlight the lack of a territorial nexus and a commitment of resources in the host state (Bayview v. Mexico<\/a><\/em> [103]). On the other hand, an investor might insist that its intangible assets, such as AI datasets and algorithms, involve the commitment of resources as their development entails the expenditure of time, capital, and human effort. In some cases, the investor may seek to demonstrate that the services it performs, even though purely digital, positively contribute to the economy of the host state (Abaclat et al v. Argentina<\/a> <\/em>[374], [378]; Ambiente Ufficio et al v. Argentina<\/a><\/em> [499]). An AI enterprise may also seek to counter the state\u2019s argument by pointing to other connecting factors, such as having a liaison office in the host state (SGS v. Philippines<\/a><\/em> [111]-[112]; Inmaris Perestroika v. Ukraine<\/a><\/em> [124]) or its activity being subject to the laws of the host state.<\/p>\n\n\n\n By way of example, a state may ban the use of a general-purpose AI platform, as a result of which the AI enterprise will lose market access in that jurisdiction. If the AI entity does not have a physical presence (such as data centres and server installations) in the territory of the host state, has not concluded a state contract, or has not obtained a licence authorising activities in the target state, it would be challenging to elevate what could be characterised as a mere cross-border provision of services to a protected investment satisfying the Salini<\/em> test elements and the territorial nexus requirement. In this context, the activity of a general-purpose AI may not entail a \u201csubstantial commitment of resources\u201d on the part of the tech company that would expose its economic interests to the regulatory and administrative \u201crisks\u201d inherent in the host state\u2019s legal environment. The collection of subscription fees by the AI entity merely generates a revenue stream for the AI company, which, if not reinvested in the territory of the host state, may not constitute assets committed in the host state and capable of qualifying for treaty protection. Furthermore, unlike some pre-shipment inspection cases (e.g., SGS v. Pakistan<\/a><\/em> [136], SGS v. Philippines<\/a><\/em> [99]-[106], Bureau Veritas v. Paraguay<\/a><\/em> [103]-[104], and SGS v. Paraguay<\/a><\/em>) where, despite the lack of substantial local presence, investment was found to exist because the economic impact of the investment was felt in the recipient state\u2019s territory, in such AI cases, any benefit to the host state is at best indirect and may be insufficient to demonstrate a \u201ccontribution\u201d to the recipient state\u2019s economy within the meaning of the Salini<\/em> test and satisfy the territorial nexus requirement.<\/p>\n\n\n\n The respondent may claim that, even if characterised as an investment, the establishment of an AI investment was not in conformity with the laws of the host state, thereby failing to satisfy a requirement of subject-matter jurisdiction.<\/p>\n\n\n\n Apart from breaches of general laws of the host state that may render an investment unlawful (such as anti-bribery laws (e.g., Metal-Tech v. Uzbekistan<\/a><\/em>)), violations of sector-specific regulatory frameworks (directly or closely related to AI) at the time of the initiation of the investment could give rise to successful illegality objections. The following are some conceivable examples:<\/p>\n\n\n\n Faced with such legality objections, an ISDS tribunal must assess:<\/p>\n\n\n\n Given that AI-related laws and regulations remain in a fledgling state across various jurisdictions, their binding force and position within the domestic legal hierarchy may be uncertain. This would complicate the arbitral tribunal\u2019s examination of whether such rules qualify as \u201cfundamental\u201d for the purposes of compliance with the legality requirement.<\/p>\n\n\n\n
Reza Eftekhar<\/a><\/strong>, Senior Legal Adviser at the Iran-United States Claims Tribunal (\u201cIUSCT\u201d)<\/strong><\/p>\n\n\n\n
\n\n\n\nThe General Legal Framework for Ascertaining Subject-Matter Jurisdiction in ISDS<\/strong><\/h2>\n\n\n\n
\n
The Existence of Rights or Interests in Property Constituting an \u201cInvestment\u201d<\/strong><\/h2>\n\n\n\n
The Existence of an Investment<\/strong><\/h2>\n\n\n\n
The Subjective Test<\/strong><\/h3>\n\n\n\n
The Objective Test<\/strong><\/h3>\n\n\n\n
Legality of the Investment<\/strong><\/h2>\n\n\n\n
\n
\n