Abstract. The convergence of Artificial Intelligence and blockchain technology has introduced novel paradigms in dispute resolution, challenging traditional frameworks for arbitration. This paper examines the enforceability of algorithmic adjudication clauses within Blockchain Dispute Resolution (BDR) platforms (specifically Kleros) under the Convention on the Recognition and Enforcement of Foreign Arbitral Awards 1958 (the New York Convention). Through doctrinal analysis of Articles I, II, IV, and V, this study addresses three fundamental questions: whether arbitration agreements embedded in smart contracts satisfy the "agreement in writing" requirement; whether parties may validly consent to non-human adjudication; and whether AI-rendered decisions qualify as enforceable "awards" despite the absence of a conventional lex arbitri. The analysis reveals that while digital consent mechanisms align with contemporary interpretations favoring in favorem negotii, significant challenges persist regarding procedural fairness, particularly the algorithmic "black box" opacity and the right to be heard under Article V(1)(b).
This analysis explores the potential future enforceability of algorithmic adjudication clauses should blockchain dispute resolution platforms evolve to incorporate autonomous AI arbitrators. While such systems do not yet exist in mature form, this analysis identifies legal prerequisites for New York Convention compatibility and proposes necessary safeguards. Proposed solutions include Human-in-the-Loop (HITL) architectures and ERC-8004 identity protocols to ensure transparency and due process. The paper concludes that algorithmic adjudication clauses, though challenging conventional boundaries, remain consonant with the New York Convention's flexible framework when properly structured and limited to appropriate dispute categories.
Key Words. Blockchain Dispute Resolution; Decentralized Justice; New York Convention; Artificial Intelligence, Arbitration.
I. INTRODUCTION
Artificial Intelligence ("AI") has achieved unprecedented levels of integration in contemporary society. Although still in the nascent stages of widespread adoption, consumer-facing Large Language Models ("LLM") have become indispensable across a diverse range of professional environments (Swain, 2025).
From individual assignments to enterprise-level implementation, AI chatbots such as OpenAI's "ChatGPT," Google's "Gemini," and Anthropic's "Claude" have fundamentally transformed the professional paradigm of the modern era (Swain, 2025). Furthermore, with the evolution of these models from mere chatbots to fully-fledged agents, the automatization of not only word generation but also entire workflows has begun to establish itself as the standard in enterprise integration.
Organizations are observing not merely financial savings, but a significant enhancement of operational efficiency. Recent studies estimate productivity increases of up to 37% (in time reduction) in professional writing and analytical tasks undertaken by mid-level college-educated professionals, including marketers, consultants, and data analysts (Noy & Zhang, 2023). Now, with the rise of artificial intelligence agents and the automation of workflows through the integration of everyday technological tools complemented and fully dominated by AI.
More recent data, as demonstrated in Papadopoulos et al. (2025), illustrates a wider range of applied use cases, reporting not only enhanced productivity but also specific examples such as a 40% reduction in inspection errors within quality control, a 30% increase in productivity through process optimization, a 25% decrease in inventory costs within Supply Chain Management, and a 35% improvement in hiring accuracy within human resources.
Major technology corporations, such as Amazon.com, have reported fiscal conservation reaching USD$260 million, effectively saving the equivalent of 4,500 developer-years of labor that would otherwise have been executed by human personnel (Cholkar, 2024). The magnitude of this integration is such that traditionally conservative sectors, notably the legal industry, are rapidly incorporating these technologies.
As stated in Martinez (2025) "courts and administrative bodies across jurisdictions are increasingly experimenting with artificial intelligence (AI) to triage filings, surface precedents, assess risk, draft summaries and—even more controversially—shape judicial outcome". Regarding the private sphere of law, "predictive analytics can aid in assessing litigation risks and settlement probabilities, improving decision-making for litigants and judges alike" (Khokhar et al., 2025). Although the resultant surge in adaptation has focused specifically on the role of AI in the field of law, this is not the sole technology to have permeated the legal sphere.
Blockchain Technology constitutes one of the recent decades' most promising innovations demonstrating significant potential for the evolution of trust and security in traditional commerce through its distributed ledger attribute, thereby introducing various proposals for integration and problem-solving solutions across society. Since its initial rise to prominence following the creation of Bitcoin and the publication of its seminal whitepaper in 2008 (Nakamoto, 2008), this technology has, approximately two decades later, expanded beyond the mere transfer of digital assets to encompass the utilization of smart contracts and the development of numerous decentralized applications (DApps) operating within the emerging and expanding Web3 ecosystem.
Blockchain Technology has now permeated multiple industries, including supply chain management, financial services, data management, the internet of things, healthcare, e-commerce, agriculture, industrial work, military and defence, and governance, as explored by Mohammed et al. (2024) and Abrar & Sheikh (2024). This technological integration has also become widespread within sectors such as the private arbitration legal industry through its incorporation into, and subsequent formation of, the evolution of Online Dispute Resolution (ODR) by means of the currently developing, rapidly expanding projects utilizing emerging protocols of Blockchain Dispute Resolution (BDR), which advocate for online decentralized justice.
This progression is demonstrated by the continually expanding body of research over the past decade, wherein the literature on Online Dispute Resolution (ODR) has increasingly incorporated discussions of Blockchain Dispute Resolution (BDR) projects. This engagement ranges from seminal works, such as Koulu (2016), which examined the initial implementations of smart contract self-execution mechanisms (if-then decisions) and their potential utility in ODR, to more sophisticated proposals like that of Kaal & Calcatera (2017). The latter introduced developing ideas of decentralized justice as "Distributed Jurisdiction" and initially posited that the "nature of smart contracting necessitates a crypto dispute resolution mechanism".
Subsequently, Schmitz & Rule (2019) further advocated for the integration of emerging blockchain dispute resolution start-ups into the settlement process for smart contracts through the explicit encoding of Online Dispute Resolution (ODR) clauses within the smart contracts themselves. At this juncture, a proliferation of new projects was underway, with Kleros notably emerging as the most prominent to date. More contemporary literature, including contributions from Allen et al. (2019), Aouidef et al. (2021), Zhuk (2023), and Gabuthy (2023), continues to explore the refined concepts of decentralized justice for blockchain smart contracts, encompassing both on-chain and off-chain methodologies.
Within the realm where law intersects with technology, Kleros has positioned itself as the most most relevant Blockchain Dispute Resolution (BDR) protocol, finding itself within important conversations regarding the technology in the use of private arbitration (Allen et al., 2019; Chen, 2022; Jovanović, 2023). Operating as a decentralized application primarily on the Ethereum platform, with plans to migrate and a current beta deployment on Arbitrum, it functions as a third-party arbitrator for contractual agreements of varying complexity.
Based on the implementation of game theory to incentivize jurors by collateralizing funds through proof-of-stake in smart-contracts, Kleros ensures the accuracy of its rulings (Lesaege & Ast, 2019). The dispute resolution mechanism employed by Kleros, primarily focused on disagreements related to smart contracts, exhibits a marked divergence from the established structure and inherent characteristics of conventional international arbitration (Santana Galindo, 2020). Notwithstanding this distinction, Kleros case uses have proceeded with a migration from its initial decentralized economy law (Web3-law) to traditional legal frameworks (TradLaw), whether it is private, commercial or even consumer arbitration, as seen in Fernandez Tineo (2019).
Another application has been the utilization of the technology within the public administration. Rather than operating as self-contained crypto-based judicial bodies as posited by certain scholars such as in Marenco et al. (2025), prior studies have positioned Kleros as a possible utility to judicial workload, particularly in the Argentinian legal scenery as seen in Fernández Tineo (2019). The fundamental hypothesis of Fernández Tineo (2019) would later be tested as Kleros's crowdsourced juror decision-making would generate significant applications and research (Brusco, 2025), particularly within the Argentine legal jurisdiction. Specifically, a 2024 pilot program conducted in collaboration with the Judiciary of Mendoza Province utilized the blockchain-based jury systems for low-complexity, off-chain civil disputes. These resolutions emitted by the Kleros jurors functioned in a non-binding and purely consultative capacity, consistently generating rulings highly aligned with the determinations of the presiding judge appointed to the case (Brusco, 2025).
The incorporation of Kleros into TradLaw frameworks (whether as ad-hoc cyber arbitration courts, judicial public administration aid, or regarding the nature and viability of its arbitral awards) has garnered significant scholarly attention, particularly concerning the recognition and enforcement of Kleros awards under the Convention on the Recognition and Enforcement of Foreign Arbitral Awards 1958 (hereinafter "the New York Convention" or the "Convention").
Scholarly works, such as those by Féliz Guerrero and Féliz Guerrero (2025), Lowther (2020), Jovanović (2026), Santana Galindo (2020), and Treacy (2022) have, to varying degrees, posited the potential for Kleros decisions to be considered arbitral awards under the Convention. Adopting a nuanced perspective, Prawira and Lewiandy (2025) contend that while Kleros generally adheres to established fairness criteria, its arbitral awards currently lack the straightforward enforceability afforded to traditional awards under the Convention. Similar to Prawira and Lewiandy (2025), Virues (2021) has argued that Kleros, rather than treated as arbitration, should rather be seen as a contractual dispute resolution mechanism.
Conversely, authors such as Pinheiro (2025) have contended an outwards negative position that Kleros awards—or rather, the Kleros protocol as a whole—should not be considered arbitration due to its lack of effect as a jurisdictional act, subsequently falling outside the Convention's scope. Consequently, the Kleros protocol and its resultant awards would be classified, pursuant to our opinion on Pinheiro (2025), and more precisely, as contractual arbitration, as explicated by González de Cossío (2018).
Nevertheless, the pervasive integration of artificial intelligence within the digital economy has demonstrably extended to the legal sector as well. The arbitration field is currently engaged in a vigorous scholarly debate regarding the operational incorporation of AI technologies—specifically AI Agents, Large Language Models (LLMs), Natural Language Processing (NLP), and Generative AI (GenAI). While some academics, such as Hussain et al. (2023) and Zhuk (2026) strictly oppose the autonomous issuance of reasoned arbitral awards by AI systems, others adopt a more nuanced stance.
Scholars like Kasap (2021) argue that such autonomy is technically unfeasible without the achievement of Artificial General Intelligence (AGI) and, even if realized, remains undesirable. Similarly, authors such as Shoukat (2025) posit that AI within the domain of international arbitration possesses significant potential to optimize the entire arbitration process, encompassing both efficiency and economic considerations. However, regarding AI-generated awards, they advocate for a cautious approach, necessitating further, more profound research into the issues surrounding this subject.
Walters (2025) contends that current legal frameworks, specifically the New York Convention, are inadequate for accommodating the concept of AI arbitrators, thereby suggesting that signatory states possess a basis to refuse recognition and enforcement, necessitating subsequent legal reforms for adaptation. Conversely, Treacy (2022) submits that a technologically sophisticated and contemporary interpretation of the New York Convention would obviate the necessity for substantive amendments to accommodate AI Arbitrators. Shrehryar & Anwer (2025) provide a more thorough perspective, offering a comprehensive examination of the challenges pertaining to the enforceability of blockchain-based arbitral awards. Their exploration encompasses "traditional enforcement standards, including decentralized proceedings, anonymity of jurors, classification of the nature of arbitral awards, formal validity requirements, arbitrability and public policy consideration" (2025).
A more affirmative stance is adopted by scholars such as Qadri (2025), along with the developers of Kleros, who advance the frontiers of this technology by championing "agent jurors" and AI-generated awards within decentralized, ad-hoc cyber courts, whether it be for On-Chain or Off-Chain simple disputes (Ast & Poorhashemi, 2026).
This paper evaluates the potential enforceability of algorithmic adjudication clauses, facilitated by Blockchain Dispute Resolution (BDR), within the framework of the New York Convention, alongside a critical assessment of potential counter-arguments.
This analysis does not aim to determine the present capacity of Artificial Adjudication to resolve disputes but rather to ascertain whether the New York Convention's framework accommodates it. The subsequent examination will meticulously scrutinize three principal dimensions: First, whether arbitration agreements incorporating AI-driven decision-making systems (such as those potentially utilized by the Kleros protocol) satisfy the formal requirements of Article II, particularly the "agreement in writing" prerequisite and the substantive question of whether parties may validly consent to non-human adjudication. Second, whether decisions rendered through algorithmic adjudication can be properly characterized as "awards" within the meaning of Article I of the Convention, notwithstanding the absence of a conventional lex arbitri and the a-national nature of blockchain arbitrations. Third, whether such awards face insurmountable barriers to recognition and enforcement under Article V, with specific attention to procedural fairness concerns including the right to be heard, algorithmic opacity (the "black box" problem), and potential public policy violations. The analysis further distinguishes between disputes of varying complexity and proposes technological solutions, including Human-in-the-Loop (HITL) systems and ERC-8004 identity protocols, to address fundamental due process concerns while preserving the efficiency advantages of decentralized dispute resolution.
II. THE VALIDITY OF THE ALGORITHMIC ARBITRATION AGREEMENT
2.1. Defining the "Agreement in Writing" in the Age of Smart Contracts.
As Redfern and Hunter have well stated "the agreement to arbitrate is the foundation stone of international arbitration." (Redfern and Hunter, 2015). This matter, undoubtedly, has prevailed as one of the most contentious issues within the field of arbitration, more precisely, the requirement that arbitration clauses be concluded in writing, including the scope and resulting implications of this stipulation.
This said, however, it would not be possible to refer to this point without having first explored the two principles upon which the contractual nature of arbitration rests: i) consensuality and ii) binding force, both of which, in their proper measure, legitimize the teleological intent of this requirement.
Regarding the former, consensuality implies the perfection of the contract, the perfect consolidation of both wills. With mere consent, the contract is formalized. Likewise, referring back to the basic principles of the contractual institution, this consent must be free and intrinsic to the party giving it, devoid of external facts that could influence or taint its purpose, without prejudice to the fact that its manifestation, in turn, may be express or tacit (Saldarriaga, 2008), or as a consequence of a legal presumption (Caivano, 2006).
It is also useful to distinguish in this section the ad substantiam formalities from the ad solemnitatem and ad probationem formalities, given that the former require a juridical event that in turn entails a legally authorized protocol for such purposes. As has been recognized by international jurisprudence, form in these cases is substance, such that legal transactions or acts do not exist as such if they are not executed under the legally prescribed form (Waste Management v. United States, arbitral award of 2/6/2000), whereas ad probationem formalities are only required for purposes of proof of the juridical event, without their absence conditioning its validity or effectiveness. This distinction is crucial to understanding the consequences of the failure to observe their formalization.
On the other hand, the principle of binding force entails the requirement that the wills of both parties concur in accordance with the formal requirements imposed by each legal system and necessary to constitute valid and enforceable sources of obligations. The corollary of its fulfillment is two-dimensional: (i) the positive effect, which implies the commitment of both parties to submit their dispute to arbitral tribunals by virtue of the contract, as well as its invocation to initiate the arbitral procedure (Saldarriaga, 2008); (ii) on the other hand, the negative effect, which entails the necessity of inhibition by the local jurisdictions (Fouchard, 1999).
Attending to these premises, in an attempt to alleviate the plural differences derived from the various national legislations and with the resistance of some drafters, the Drafting Commission opted to include the brief annotation that the agreement had to be "in writing" (Saldarriaga, 2008). However, far from what the Convention might deceptively suggest, this writing requirement was interpreted to denote "an arbitral clause in a contract or an arbitration agreement, signed by the parties or contained in an exchange of letters or telegrams".
As Saldarriaga (2008) rightly maintains, the intention of the delegates when drafting the Convention was to preserve the formality of this requirement at all costs, endowing it with an ad substantiam character, which means that, from the outset, it was foreseen that the clause be contained in some documentary support that would allow accrediting the consent of both parties. For the drafters of the Convention, it would not be enough for one party to present a contract containing an arbitration clause; rather, it must be duly signed, so that consent is unequivocal and sufficiently transparent.
The foregoing was equally supported by Italian national courts which expressly held that the written form is required ad substantiam by virtue of Article II of the Convention, and that this requirement is satisfied in cases where the arbitration clause is set forth in a contract, an agreement signed by the parties or contained in an exchange of letters or telegrams (Gaetano Butera v. Pietro Romano Pagnan, 1978). This treatment suggests that this clause has been treated by some national jurisdictions in an excessively formalistic manner and, consequently, the transactional and commercial reality of the day-to-day has been disregarded.
Nevertheless, a review of international jurisprudence from various national courts over the past five decades reveals certain inconsistencies. Notwithstanding the teleological intent underlying the clause's initial conception and the interpretation some tribunals might apply in isolation, it has progressively been recognized that the consent of a party may be tacitly granted and inferred from their conduct throughout the commercial relationship, for instance, through their participation in the contract's execution (Saldarriaga, 2008). Thus, the requirement of writing could be satisfied with the mere presentation of a document that rendered the existence of the agreement incontrovertible, or the exchange of documents without the need for the signature of one of the parties.
This radically and transversally changed the initial formalist paradigm of these clauses (Landau, 2003). Legal doctrine and national courts have supported this stance, thus reinforcing that the requirement of writing does not truly have an ad substantiam character, allowing, instead, a glimpse of its ad probationem nature. This criterion is further reinforced when considering that Article IV.1 (b) of the same Convention requires the presentation of the agreement in original or copy to prove the existence of its content (Saldarriaga, 2008).
These assertions necessitate the specification of two crucial caveats: i) the requirement of a written agreement for evidentiary purposes under Article IV, where an indication of the initial intent to arbitrate is deemed sufficient; and ii) the necessity of this verification would apply solely to the recognition and enforcement phase, such that a national judge, when seized of a referral request, is not bound by the conditions stipulated in Article IV. Rather, the judge against whom the recognition and enforcement of the award is sought, could refuse to recognize it due to the non-observance of Article IV.
In short, we can conclude that the Convention's requirement that the arbitration agreement be in writing does not constitute a formality ad substantiam, and that it would only be required ad probationem for the eventual recognition and enforcement of the award, with this requirement being satisfied by the mere presentation of a document confirming the existence of an intention to arbitrate.
Nevertheless, the central challenge for this section lies in determining whether these considerations apply to novel, technology-driven arbitration clauses formulated to satisfy increasing enterprise requirements. The growing utilization of crypto-arbitration within enterprise dispute resolution frameworks, particularly its valued attributes of celerity and efficiency in meeting the specific requirements of electronic commerce, necessitates a thorough examination of its enforceability.
First and foremost, it should be clarified that the platforms resorting to Blockchain Dispute Resolution (BDR) do so through "arbitration clauses," whether it be inserted into smart contracts or in writing as seen in Lowther (2020); which may, in turn, concern either off-chain or on-chain disputes. More specifically, these contracts can—and often do—incorporate the customary traditional dispute resolution clauses, such as the arbitration clause—also referred to as submission clauses—which stipulates the parties' agreement to resolve conflicts arising from blockchain operations, and for these decisions, in turn, to be automatically executed on the same blockchain (Bergolla et al., 2022) through platforms like Kleros.
Given this scenario, the question arises as to whether a clause, conceived under these standards and specially designed to meet the emerging needs of electronic commerce, could be recognized in light of the Convention. Authors such as De Cossío have suggested that, within the framework of electronic commerce, an arbitration agreement is understood as valid if it is (i) in writing: meaning, contained in a data message that is integral and accessible for subsequent consultation; and, (ii) if it is signed, if it allows for the individualization of the signatory and it can be verified that the latter approved the information contained in this data message (De Cossío, 2018). This last requirement serves the clear purpose of verifying and confirming the signatory's consent.
Thus, an analogous and extensive interpretation of the foregoing suggests that these arbitration clauses, whether embedded within smart contracts or contained in documentary supports (in the case of an off-chain dispute), satisfy the fundamental requirements for validity. For smart contracts, these clauses are accessible on the blockchain for subsequent review and are digitally consented to upon the conclusion of negotiations within this ecosystem, where transparency and verifiability of information are maintained consistently. "The utilization of a smart contract as a contractual instrument presupposes the parties' agreement to the stipulated contractual clauses" (Pinheiro, 2025).
In this same vein, when parties agree to submit disputes to Kleros ODR, their consent is formalized and carried out through a smart contract. As a result, there is a clear and verifiable record of their agreement to use Kleros ODR, thereby satisfying the New York Convention's requirement that an arbitration agreement be in writing (Lowther, 2020).
In addition to the foregoing, UNCITRAL, aware of this undeniable reality, have undertaken many efforts to relax the form under which the agreement can be formed (Redfern and Hunter, 2015). Example of this was the issuance of an exhortatory communication to all nations to apply Article II(2) in the laxest manner and most favorable to the purposes of accommodating electronic and modern commerce, to recognize new business schemes in their multiple facets and presentations (UNCITRAL, 2006). In this same spirit, UNCITRAL has been inclined to suggest increasingly inclusive reforms aimed at adapting to the new needs of commerce, including the adoption of two texts as options to Article 7 of the UNCITRAL Model Law.
It follows from the foregoing that the requirement of a written arbitration agreement is not rigid but rather adaptable, reflecting a policy that favors the validity of agreements (in favorem negotii). Nonetheless, the writing must still fulfill three essential functions: it must provide evidence that the parties consented to arbitrate, clearly establish the precise terms of that consent, and demonstrate that the parties understood that choosing arbitration entailed waiving recourse to state courts (Jovanovic, 2025).
Therefore, the evidence suggests that the clauses contained in smart contracts, and from which the referral to arbitration arises, comply with the necessary elements to be considered valid. Ultimately, (i) it is in writing in the electronic terms and must be signed by the parties, which allows verification of the consent given by them; (ii) it is available for later consultation.
2.2. Algorithmic Adjudication within Legal Practice and Arbitral Decision Making.
Before determining whether parties may validly consent to such clauses, it is necessary to establish what "algorithmic adjudication" entails and how it differs from traditional arbitral decision-making.
Within the legal sphere, the introduction of AI algorithms has permitted what is known as algorithmic adjudication, where "decision-making is shaped not only by legal standards but also by computational logic embedded in platform infrastructures" (Reddy et al., 2023). This is achieved by "introducing computational models capable of mimicking (though not yet replicating) some dimensions of legal reasoning" (Espinal de Aza, 2025). Mostly, algorithmic adjudication has been used for predictive justice, using algorithms to, among other functions, "identify disparities in sentencing, reduce human bias, and optimize resource allocation" (Espinal de Aza, 2025). This perspective suggests that "algorithmic decision-making may turn out to be faster and cheaper than human decision-making, but those are only virtues if the decisions rendered are reasonably accurate" (Chiao, 2024).
A wide variety of literature has, however, expressed opposition to and cautioned against the integration of Algorithmic Adjudication with the rule of law, or most importantly, "Human-out-of-the-Loop" (HOOTL) systems. Therefore, the necessity of incorporating a "Human-in-the-loop" (HITL) methodology has been a consistent assertion within this pertinent academic discourse (Espinal de Aza; 2025; European Commission for the Efficiency of Justice, 2018; Khokhar et al., 2025; Martinez, 2025). These instruments are officially characterized as decision-support mechanisms, rather than substitutes for judicial authority (Martinez, 2025).
This position has been substantiated by clear case studies, such as the instance in Brazil involving the development of a tool utilizing Generative AI for drafting judicial decisions (Martinez, 2025). As a result, several academics, such as Espinal de Aza (2025), maintain that the technology remains underdeveloped, asserting that, "given the present state of technology and the normative demands of the legal system," fully autonomous decision-making is currently unfeasible. Espinal de Aza (2025) argues that, "as these systems learn from historical data, they inevitably reproduce historical inequalities. Should the past contain inherent bias, the model institutionalizes it. The algorithm, therefore, functions as a reflection of structural injustice—a reflection mistakenly perceived as objective truth."
While the prevalent legal-technological discourse generally favors a conservative approach, a more progressive viewpoint exists within the realm of Algorithmic Adjudication, frequently characterized by proponents of "AI Judges". Volokh (2019) advanced the futurist possibility concerning the development of these AI-powered judges (HOOTL) before the rise of Generative AI, which is the technology now extensively employed to facilitate AI decision-making. "Human judges, though being human, have human prejudices"..."leaving decisions, or at least certain kinds of decisions, entirely to AI judges may help avoid these prejudices" (Volokh, 2019). Fundamentally, "the basic criterion for promotion should still be whether we trust the candidate's judgment," suggesting an implicit acknowledgment that, if the judgment rendered by Artificial Intelligence is sufficiently reliable, there should be no impediment to trusting its reasoning, thus favoring HOOTL systems (Volokh, 2019).
The conclusion made in Volokh (2019) suggests that arbitration is an ideal sector for developing and testing this technology, given the inherent needs for efficiency and cost management. Furthermore, the author advocates for applying the technology even to sectors, such as consumer affairs, that currently exhibit a de facto bias toward one of the parties, positing that this type of bias would be more difficult for Artificial Intelligence to harbor.
The inherent challenge, as elucidated by Volokh (2019), is that the optimism surrounding these systems—specifically their potential for independence and lack of external supervision—is entirely contingent upon their actual capacity to operate without oversight. This necessitates that the systems generate trustworthy outputs, adequately substantiate their reasoning, and successfully pass rigorous, specific testing. Legal-technological discourse regarding Artificial Intelligence is replete with such optimistic projections, many of which have proven to be overly sanguine when evaluated against empirical reality. Cohen-Sasson (2025) investigates the practical outputs of current Generative AI tools, revealing a reality that starkly contrasts with the optimistic portrayals of "AI Judge" HOOTL systems. Cohen-Sasson (2025) "unequivocally demonstrates that AI systems, specifically LLMs, exhibit substantial inconsistency in their outputs—producing divergent legal conclusions across tasks and models."
Despite current experiments suggesting a clear incompatibility with the advanced future of AI Judges, Cohen-Sasson (2025) identifies two critical factors: for complex legal disputes, AI evidently fails to meet a quality standard; however, in uncomplex legal disputes, "performance improves dramatically to 89% and 93%." This is further substantiated by Datzov (2025), which establishes that "if the standard is merely the capability to issue a decision that resolves a disputed legal issue in an acceptable way, AI's technical ability may already exist or is likely to be imminent." The arguments presented by Cohen-Sasson (2025) are largely shared with Kleros (2026a), who posit that subjective and complex cases are better suited for traditional resolution methods, while a middle ground of cases is more appropriately handled by Decentralized Justice, and a significant use case for AI-powered dispute resolution exists for low-to-mid-level objectively simple cases.
In the context of Blockchain Dispute Resolution (BDR), particularly when analyzing platforms such as Kleros and the application of Human-Out-of-the-Loop (HOOTL) decision-making, the foundational protocol employs Game Theory principles to regulate the conduct of jurors (both human and artificial intelligence participants) by incentivizing them to converge upon the Schelling Point. It is essential to acknowledge that the anonymity of jurors, coupled with the absence of validation requirements for decision-making within Kleros, may give rise to contention. As Kleros does not prioritize the verification of the jurors' true human nature, it potentially accommodates a diverse array of entities serving as adjudicators.
Fundamentally, the outcomes in Kleros are driven not by conventional human judgment but by the mechanics of game theory itself. Outside of the justice protocol, the Kleros team has been developing "Proof-of-Humanity," a system where verified human participants can receive Pinakion (PNK) for confirming their authenticity (Kleros, 2026b). If integrated into the Kleros Protocol, this feature would enable users to contractually specify whether their jurors must be verified humans, thereby bridging the crucial information gap concerning the identity of the juror as either a sentient human being or an AI Agent.
As will be further elaborated, Kleros does not currently feature an AI-powered court, with the exception of a beta curation court. Consequently, apart from the aforementioned court, there is no other platform that inherently incentivizes or necessitates the employment of AI jurors. Nevertheless, this circumstance, as previously stated, does not preclude human participants from utilizing AI to substantiate their decisions or to integrate any AI tool into the decision-making process. Such practice is not actively encouraged nor discouraged, as the staked PNK required for participation in a court can be sanctioned if the resulting decision fails to achieve the Schelling point for the case (Ast, 2020; Lesaege & Ast, 2019).
Regarding AI agents, any agent is presently permissible, given that the juror drawing pool does not discriminate based on whether, as previously explained, the user is enrolled in the Proof of Humanity (PoH) program. Should Kleros choose to implement such a distinction, they would also possess the capability to do so for AI agents, facilitated by the expanding digital AI economy and the deployment of ERC-8004, which enables agents to establish an identity within blockchain technology (Hu & Rong, 2025).
3.2.1. Subjective Arbitrability: Can Parties Validly Consent to Non-Human Adjudication?
Given that interpretations of Articles II(1) and II(2) of the New York Convention, as advanced by UNCITRAL and international jurisprudence, appear to countenance an extrapolation that could accommodate Kleros arbitration in fulfilling this requirement, the assessment must now pivot to the substantive issue: the enforceability of so called "Algorithmic Adjudication Clauses".
As articulated by Shih & Chang (2024), and in congruence with the preceding discussion, "arbitration agreements are fundamentally contractual in nature; therefore, parties should generally possess the autonomy to contractually select Artificial Intelligence to serve as the arbitrator for the resolution of their disputes." Consequently, it can be understood that Algorithmic Adjudication Clauses would merely necessitate the mutual assent of both parties, thereby presenting no issues concerning party autonomy. "In contrast to conventional arbitration, where territorial sovereignty dictates the applicable legal framework, AI arbitration primarily relies on the principle of party autonomy, whereby the parties designate the governing law of the arbitration process" (Daradkeh, 2025).
In this regard, according to De Cossío (2018), consent as the central element of arbitration is understood as the concurrence of wills regarding the creation or transfer of rights or obligations, and, when extrapolated to the arbitral sphere, as the agreement between the parties to submit a present or future dispute to arbitration. Under this premise, the validity of the agreement requires only that (i) the parties possess full legal capacity to enjoy and exercise their rights; (ii) their consent is free from voices of consent; (iii) the purpose of the contract is lawful; and (iv) the formal requirements applicable to the agreement are satisfied (De Cossío, 2018), as discussed in the previous section. From this perspective, there is no express impediment to the parties' authority to agree to the use of Non-Human Adjudication for the resolution of their disputes within Kleros as a Decentralized Justice protocol.
Beyond the technical aspects inherent in Kleros as a Decentralized Justice protocol, the primary concern regarding the validity of such clauses extends outside the scope of the New York Convention and aligns with public policy considerations (Shih & Chang, 2024). The majority of arbitration laws, including the UNCITRAL Model Law and the New York Convention itself, were drafted under the presumption that an arbitrator constitutes a "natural person" (a human being) (Martinez, 2025; Shih & Chang, 2024; Shoukat, 2025); however, more modern interpretations, ranging from doctrinal positions (Walters, 2024) to evidentiary support (Sienicki & Sienicki, 2025), are emerging. This necessitates, as expounded by Daradkeh (2025), that the parties engage in forum shopping rather than being preoccupied with public policy concerns.
2.2.2. The fusion of Algorithmic Adjudication and Blockchain Dispute Resolution: Current Status and Prospective Trajectories.
The current structure of crypto-arbitration, which utilizes Artificial Intelligence, diverges somewhat from conventional expectations. Rather than the parties selecting specific AI models for dispute adjudication, individual jurors (assigned cases through Kleros's standard protocol) independently elect to utilize AI models at their discretion to accelerate decision-making. This practice reduces typical resolution times from one to three days to approximately sixty seconds (Ast & Poorhashemi, 2026). This constitutes a unilateral determination by jurors who employ AI to automate the identification of the Schelling Point, which is intended to yield the most equitable resolution in coordination with their anonymous co-jurors.
Critically, this system does not constitute algorithmic adjudication in a pure HOOTL sense, nor does it resemble automated awards in the strict sense. Rather, it operates as a HITL system where AI automates the processes of evaluating evidence, analyzing information, and generating decisions based on both the LLM's training data and case-specific inputs, but with human input and ultimate control.
However, this description reflects aspirational possibilities more than current reality. In discussing algorithmic adjudication clauses, we are effectively anticipating developments beyond Kleros's existing experiments and even beyond how its sole AI-incentivized court currently functions. Kleros's documented AI testing has been limited to simple, objective decisions, such as evaluating football match outcomes using commercial AI models (Ast & Jackson, 2018). The single court that actively encourages AI use is designed for straightforward, objective identification tasks leveraging the model's neural network capabilities.
For algorithmic adjudication clauses to function effectively within Kleros's protocol, two approaches could be implemented: I) Specified AI-HITL Model: Parties could contractually specify that jurors must employ particular AI systems connected to the adjudication mechanism, maintaining the HITL framework while standardizing the technology used; or, II) Dedicated AI Court: A specialized court could be created where jurors provide API keys for designated AI models or registered ERC-8004 agents, enabling the court to adjudicate disputes autonomously. This structure does not currently exist and would require substantial protocol development.
In practice, while parties can theoretically select both Kleros as their dispute resolution mechanism and AI as their adjudication method, unifying these choices presents complications. As established, no Kleros court currently permits AI-based decision-making outside the single specialized Automated Curation Court. To respect party autonomy fully, Kleros must develop additional protocol infrastructure within its ecosystem. This development should be modular, granting contracting parties granular control over AI configuration, including specifications such as: i) Minimum context window requirements; ii) Training data characteristics; iii) Model parameter thresholds (e.g., requiring high-parameter models); iv) Commercial versus open-source model preferences; and, v) Specific model providers or versions. Only through such structured development can the protocol genuinely accommodate AI adjudication while maintaining the principle of party autonomy central to arbitration frameworks.
III. CHALLENGES TO RECOGNITION AND ENFORCEMENT OF BLOCKCHAIN ARBITRAL AWARDS
3.1. Can Kleros Awards Rendered By AI be Considered Awards in the Sense of the NYC?
Having clarified the possibility of agreeing to the use of AI as potential adjudicators in decisions rendered within Kleros, it becomes necessary to ask whether the resulting awards may be considered as such under the Convention and, consequently, be enforceable. A crucial question in this discussion is whether Kleros awards rendered by AI can be regarded as "awards" within the meaning of the Convention. In this context, Article I(1) assumes particular significance, as it constitutes the foundation upon which the Convention is built by clarifying which arbitral awards fall within its scope of application. Specifically, it covers awards made in the territory of a State other than the one where recognition or enforcement is sought, as well as awards that are not considered domestic under the law of the State where recognition is requested. Both categories are thus treated as "foreign" awards for the purposes of the Convention (Bagner et al., 2010).
The etymological content of an arbitral award has been another heavily debated issue in arbitration scholarship. Although no universally accepted definition exists, the closest approximation is found in Article I(2) of the New York Convention, according to which the term "award" encompasses not only decisions issued by arbitrators appointed for a particular case, but also those rendered by permanent arbitral bodies to which the parties have agreed to submit. This definition is particularly relevant when considering Kleros as a crypto-arbitration protocol. According to the abstract of the Kleros White Paper, Kleros is a decentralized application built on Ethereum that functions as a neutral third party for the arbitration of contractual disputes (Lesaege & Ast, 2019). The consensual element arises through the smart-contract framework in which the parties designate Kleros as their dispute-resolution mechanism (Lowther, 2020).
Legal scholarship has identified additional elements typically required for a decision to qualify as an arbitral award: (i) it must be rendered by arbitrators; (ii) it must resolve a disputed issue; (iii) it must address the merits rather than a purely procedural matter; (iv) it must be binding on the parties; and (v) it may be partial in nature (De Cossío, 2018)
Determining whether decisions issued by Kleros protocol tribunals qualify as arbitral awards carries significant practical consequences for their operation and effects. As noted by De Cossío (2018), if a decision is considered an award: (i) the formal requirements governing the issuance of awards must be observed; (ii) the procedural steps established in the applicable arbitration rules must be followed; (iii) the decision is binding and mandatory upon the parties; (iv) it produces res judicata effects; (v) its issuance triggers the time limit for potential nullity actions and (vi) only a decision recognized as an award may be recognized and enforced under international conventions.
That said, the central question of this study, therefore, is whether decisions rendered by AI could be considered awards within the meaning of the New York Convention. A close reading of the Convention's provisions is essential to address this issue, yet it quickly gives rise to doubts. The Convention conceives an award as a decision rendered either by arbitrators or by a permanent arbitral tribunal (Art. I(2)). Even in the latter case, an arbitrator must be appointed and the parties notified (Art. V(1)(b)). Concepts such as "arbitration proceedings" and "arbitral procedure" appear to presuppose human interaction, traditionally understood as involving written or oral submissions, arguments, and deliberation among individuals. An algorithm may also face difficulties in producing a "duly authenticated award" (Art. IV(1)(a)) if authenticity is interpreted as attribution to a specific natural person (Lehmann, 2025). Similarly, the reasoning typically required in arbitral awards is uncommon in blockchain arbitration, where jurors may cast their votes independently and are not necessarily required to explain the basis for their decision. However, as the same author aptly observes, these are not insurmountable incompatibilities, since arbitral modalities such as ex aequo et bono may exist, in which a fully reasoned decision is not always a necessary element (Lehmann, 2025).
Faced with this dilemma, scholars such as Lehman (2025) take a more restrictive position, concluding that such decisions do not qualify as an "award" within the meaning of the New York Convention and therefore do not benefit from the privileges granted under that instrument. Were it otherwise, virtually any determination could be certified as an "award" and produce res judicata effects in other contracting states. Although certain settlements may circulate as arbitral awards, these are invariably preceded by proper arbitral proceedings; settlements reached in the absence of such proceedings cannot be regarded as awards. The same reasoning, in this view, applies equally to BA and AI decisions.
By contrast, other scholars, such as Jovanić (2025), adopt a far more optimistic view, arguing that issues like the requirement of a "written" arbitration agreement for the validity of the proceedings, or the need for written form and a certified copy of the arbitral award for its enforceability, do not constitute insurmountable obstacles. This is because balancing principles, such as in favorem negotii, enable interpretations that are responsive to the contemporary demands of electronic commerce and that progressively adapt to the complexities arising from the diverse forms in which arbitration agreements may be expressed.
3.2. Prospective Impediments to the Recognition and Enforcement of Kleros Awards Adjudicated by Artificial Intelligence
Another significant challenge facing arbitral awards rendered by AI is their recognition in the absence of a valid lex arbitri. Blockchain arbitrations do not have a conventional seat, which classifies their outcomes as a-national or "floating" awards. This issue becomes particularly relevant in light of Article I of the New York Convention, which provides that the Convention applies to the recognition and enforcement of awards made in a territory different from that in which recognition is sought, as well as to awards that are not considered domestic in the enforcing state. Accordingly, as Jovanić (2025) observes, an award must, as a preliminary matter, display some form of affiliation with a state other than the one where recognition is pursued.
The difficulty lies in the fact that recognition of a-national and delocalized awards such as those rendered through Kleros, may be denied ex officio for contravening public policy and for preventing the possibility of seeking annulment under any national arbitration law or jurisdiction. Although Jovanić (2025) rightly argues that the public policy clause must be narrowly interpreted and invoked only where the fundamental principles of the state are breached, it is equally true that, in the case of a-national awards, parties may be unable to invoke due-process protections under Article V(1) of the Convention. Blockchain arbitrations are often conducted with fewer procedural safeguards, prioritizing cost-efficiency and speed; thus, as the same author notes, it would be unreasonable to allow parties, when contesting recognition, to rely on procedural guarantees they initially agreed to waive.
As a safeguard, Jovanić (2025) suggests selecting an arbitral seat that is friendly to blockchain technologies and (one might add) to AI governance frameworks, since doing so could prevent potential nullities arising from the absence of a seat in blockchain arbitration. From another perspective, scholars such as Lehmann (2025), Schmitz (2025), and Zhuk (2025), examining awards rendered by large language models, categorically reject the possibility of enforcing decisions issued under such circumstances. In their view, these determinations do not qualify as enforceable arbitral awards because they lack essential elements traditionally associated with an award, such as deliberation among arbitrators or the capacity to generate res judicata effects.
A different reading suggests that the absence of a seat should not pose insurmountable difficulties in this specialized form of dispute resolution. Disputes handled through blockchain dispute resolution mechanisms do not generally follow the same degree of procedural rigor or technical formality as traditional arbitration. The more plausible concern may lie in whether the public-policy standards of the enforcing state would disapprove of decision-making mechanisms involving AI, potentially hindering enforcement. Nevertheless, where the parties have given free and valid consent, this issue should not be overstated. By opting for platforms such as Kleros, parties knowingly waive certain procedural prerogatives in exchange for a simpler and more expedited process. In a system where informality is a defining feature, it would be inconsistent to impose the same formal requirements expected in conventional arbitral proceedings.
IV. PROCEDURAL FAIRNESS AND THE CHALLENGE OF ALGORITHMIC OPACITY
Regardless of the facet of arbitration, whether it involves artificial intelligence agents, is conducted through the blockchain mechanism, or a combination of both, it is primarily governed by the conventional rules that the parties have agreed upon. However, this freedom of contract does not empower the parties to alter their procedural equality nor to undermine the constitutional guarantee of the right to defense in trial. Affecting these principles would imply a violation of the guarantee of due process, which includes, among other elements, the right to allege and to be heard (Caivano, 2011).
As Ohnoutková (2025) observes regarding AI arbitral awards, "the most relevant grounds for refusing to recognize the award with respect to the use of AI seem to be a violation of the right to be heard set in Article V(1)(b) of the New York Convention" or violations of public policy. The right to be heard in the context of algorithmic adjudication presents two fundamental challenges: (1) the failure to provide adequate notice to parties, and (2) the inability of parties to meaningfully present their case.
Most importantly, one of the major issues to be examined within BDR Algorithmic Adjudication is the inability of parties to meaningfully present their case, not because the protocol would not allow them, or because they were not duly cited to the process, but due to the technological limitations presented by the AI itself. Two of these issues will be explained below: the problem with algorithmic opacity and decision-making, and the technical problems that the context window may present. Based on this, we will present a possible solution regarding agent attribution utilizing ERC-8004.
Algorithmic opacity, frequently termed the "black box" effect, is a critical issue. Ali (2026) characterizes this opacity as resulting in "outputs that are hard to understand or explain," arising when the system "cannot explain the reasoning behind their conclusions" (Horton, 2023). Within the legal domain, this can manifest as decisions where "experts can't figure out why they were made" (Ali, 2026), which may potentially contravene principles of procedural due process (Horton, 2023). Cohen-Sasson (2025), inconsistency is a major issue confronting AI decision-making. Ohnoutková (2025) argues that this inconsistency could create issues with the same article Article V(1)(b) because the explanations may be "tailored to the AI researchers themselves, rather than effectively addressing the needs of the intended users."
To mitigate the "black box" effect from posing an issue within the framework of the New York Convention, parties would require the ability to review the outcome generated by AI. This oversight is necessary, as its absence "may create room for the application of Article V(1)(b) of the New York Convention by national courts of a contracting state" (Ohnoutková, 2025). Such mitigation necessitates transparency levels enabling parties to "understand how a particular output was generated based on specific inputs" (Ohnoutková, 2025). This would also include being able to understand or know based on what the system was trained on. This is due to AI systems only being "as sound as the data and assumptions on which they are built." (Martinez, 2025).
Regarding context windows, the reality is that it is not possible to ensure a fair process in a dispute if algorithmic adjudication is combined with complex cases. This is because artificial intelligence models are constrained by their context windows, which suggests that, if a highly complex case is presented to the AI, it may omit critical data or even fail to integrate arguments or decision documents into the final resolution due to an inability to evaluate the full evidentiary record. Therefore, I argue that while this type of conflict resolution will be implemented within Blockchain Dispute Resolution protocols, the disputes must be simple, as maintained by Kleros (2026a).
Given the evolutionary potential of artificial intelligence systems, and assuming the constraint of the context window is resolved, it becomes imperative to assign a distinct identification to the agents participating in the Kleros protocol. This necessity stems from the theoretical requirement that parties must not only consent to the decision being rendered by an AI Agent but must also—for reasons of due process and the right to defense, which are inherent constitutional considerations—be informed of the identity of said Agent. This perspective closely aligns with the experiment conducted and documented by Ast & Jackson (2018).
However, to preserve the anonymity of the jurors and prevent their mutual communication, our proposal advocates for a mechanism by which a virtual identity is conferred upon these agents through the ERC 8004 standard. This identity must, in turn, satisfy the intrinsic requirements of Kleros and comply with the conditions previously enumerated in Point III. This approach is presented as our solution to guarantee the agent identification essential for procedural fairness and compliance with the New York Convention (NYC).
V. CONCLUSION
The assessment of the enforceability of arbitral awards rendered pursuant to Algorithmic Adjudication Clauses within the Blockchain Dispute Resolution (BDR) ecosystem suggests a legal framework that is demonstrably consonant with the flexible structure of the New York Convention, contingent upon the implementation of specific interpretive modifications.
As rigorously established throughout this examination, the prerequisite for an "agreement in writing" stipulated under Article II no longer constitutes the principal impediment to decentralized justice; the maturation of digital consent via smart contracts effectively satisfies this formal requirement. The salient challenge resides, instead, in the substantive procedural safeguards mandated by Article V, particularly concerning the "Black Box" phenomenon and the fundamental right to be heard. Nevertheless, the outright rejection of Algorithmic Adjudication based exclusively upon these reservations would fail to acknowledge the nuanced reality of the specific disputes for which these mechanisms are engineered.
A crucial differentiation must be established concerning the inherent complexity of the controversy. As recent academic discourse and the developmental trajectory of platforms such as Kleros indicate, AI-driven adjudication is presently most efficacious for objective disputes of low-to-medium complexity, where the imperative for expedition and cost-efficiency of the determination supersedes the necessity for exhaustive human deliberation. In these precisely defined contexts, the application of the stringent procedural standards characteristic of traditional high-stakes commercial arbitration to BDR awards would be logically inconsistent.
Fundamentally, the enforceability of these clauses is predicated upon the cardinal principle of party autonomy. Should parties, possessing full comprehension of the protocol's operational mechanics (including the incorporation of Game Theory and AI agents) voluntarily relinquish certain conventional procedural entitlements in favor of a decentralized mechanism, their manifestation of consent ought to be duly respected. While the prevailing "Black Box" opaqueness inherent in specific AI models introduces an enforcement risk under public policy provisions, the forthcoming adoption of "Human-in-the-Loop" architectures or explicable AI agents (through protocols such as ERC-8004) presents a viable and progressive solution.
Consequently, while Algorithmic Adjudication undeniably challenges the established boundaries of the New York Convention, it does not necessarily contravene its underlying ethos. Provided that the dispute remains within the domain of subjective arbitrability and the parties have explicitly acceded to the technological parameters governing the adjudication, national judiciaries should embrace an in favorem negotii posture, recognizing these determinations not as deviations from justice, but as the inevitable evolution of dispute resolution within the digital era.
VI. REFERENCES
Abrar, I., & Sheikh, J. A. (2024). Current trends of blockchain technology: architecture, applications, challenges, and opportunities. Discover Internet of Things, 4(1), 7.
Ali, S. I. (2026). Artificial Intelligence and the Law: Human Rights, Justice Systems, and Governance in the Age of Autonomous Decision-Making. Artificial Intelligence, 2(1).
Allen, D. W., Lane, A. M., & Poblet, M. (2019). The governance of blockchain dispute resolution. Harv. Negot. L. Rev., 25, 75.
Aouidef, Y., Ast, F., & Deffains, B. (2021). Decentralized justice: A comparative analysis of blockchain online dispute resolution projects. Frontiers in Blockchain, 4, 564551.
Ast, F., & Jackson, S. (2018, June 28). CrowdVAR: Could Kleros become the football referee of the future? Collective intelligence for a transparent beautiful game. Kleros.
Ast, F. (Ed.). (2020). Dispute revolution: The Kleros handbook of decentralized justice (Revised and augmented ed.). Kleros.
Ast, F., & Poorhashemi, A. (2026, February 6). Between crowds and algorithms: Decentralized justice in the age of AI [Webinar]. Kleros; International Law Institute (CIFILE). LinkedIn.
Bagner, H. (2010). Title of the chapter. In H. Kronke, P. Nacimiento, D. Otto, & N. C. Port (Eds.), Recognition and Enforcement of Foreign Arbitral Awards: A Global Commentary on the New York Convention (pp. 19–36). Kluwer Law International.
Bergolla, L., Seif, K., & Eken, C. (2022). Kleros: A socio-legal case study of decentralized justice & blockchain arbitration. Ohio State Journal on Dispute Resolution, 37(1), 55–98.
Blackaby, N., Partasides, C., Redfern, A., Hunter, M., & others (2015). Redfern and Hunter on international arbitration (6.ª ed., p. 77). Oxford University Press.
Brusco, F. (2025). From Blockchain To Bench: Evaluating The Kleros Pilot In Mendoza's Judicial System [Report]. Kleros Fellowship of Justice.
Caivano, R. (2006). Arbitraje y grupos de sociedades: extensión de los efectos de un acuerdo arbitral a quien no ha sido signatario. Lima Arbitration, (1), 142.
Caivano, R. J. (2011). Control judicial en el arbitraje. AbeledoPerrot.
Chen, J. W. (2022). Dispute Resolution in the New Digital Era-Exploring Arbitration as a Suitable Mechanism to Resolve Disputes over Crypto Assets. Contemp. Asia Arb. J., 15, 255.
Chiao, V. (2024). Algorithmic Decision-making, Statistical Evidence and the Rule of Law. Episteme, 21(4), 1241-1264.
Cholkar, A. A. (2024, August 1). Amazon Q Developer just reached a $260 million dollar milestone. AWS DevOps & Developer Productivity Blog.
Cohen-Sasson, O. (2025). Stochastic Justice: Legal Inconsistency by Humans and Machines. Available at SSRN 5380801.
Daradkeh, L. (2025). THE APPLICABILITY OF NEW YORK CONVENTION OF 1958 ON ARTIFICIAL INTELLIGENCE ARBITRAL AWARD: EXPLORING CONTROVERSIES THROUGH MAPPING AND SEEKING INTERPRETATION TO ENHANCE COORDINATION. Scientific Culture, 11.
Datzov, N. L. (2025). AI Jurisprudence: Toward Automated Justice. Northwestern Journal of Technology and Intellectual Property, 23(1), 1.
Espinal De Aza, E. L. (2025). The Courts in the Age of Artificial Intelligence: Legal Transformation, Ethics, and Algorithmic Justice. Ethics, and Algorithmic Justice (October 01, 2025).
European Commission for the Efficiency of Justice. (2018). European ethical Charter on the use of artificial intelligence in judicial systems. Council of Europe.
Féliz Guerrero, G., & Féliz Guerrero, J. (2025). Legal viability of decentralized justice in the Dominican Republic: Application of Kleros in different economic sectors (Kleros Fellowship of Justice, 8th Cohort). Kleros.
Fernandez Tineo, J. (2019). Application of the Kleros decentralized dispute resolution technology to the Argentine Courts in cases of civil liability derived from parental responsibility (Fellowship of Justice Program 2019). Kleros.
Fouchard, P., Gaillard, E., & Goldman, B. (1999). Fouchard, Gaillard, Goldman on international commercial arbitration. Kluwer Law International.
Gabuthy, Y. (2023). Blockchain-based dispute resolution: insights and challenges. Games, 14(3), 34.
Gaetao Butera v. Pietro e Romano Pagnan. (1978, 18 de septiembre). Sentencia, Corte di Cassazione (Sezioni Unite), No. 4167. Yearbook Commercial Arbitration, 4, 296–300 (1979).
González de Cossío, F. (2018). Arbitraje (1st ed.). Editorial Porrúa. (Original work published 2004)
Horton, D. (2023). Forced robot arbitration. Cornell L. Rev., 109, 679.
Hu, B. A., & Rong, H. (2025). Inter-Agent Trust Models: A Comparative Study of Brief, Claim, Proof, Stake, Reputation and Constraint in Agentic Web Protocol Design-A2A, AP2, ERC-8004, and Beyond. arXiv preprint arXiv:2511.03434.
Hussain, M. A., Labanieh, M. F., Mahdzir, N., Sulaiman, N., & Bawazir, O. S. A. (2023). The potential prospect of Artificial Intelligence (AI) in arbitration from the international, national and Islamic perspectives. Journal of International Studies (JIS), 19(1), 95-122.
Jovanović, S. (2023). ARBITRATION IN SMART CONTRACTS DISPUTES–A LOOK INTO THE FUTURE. Анали Правног факултета у Београду, 71(4), 757-784.
Kaal, W. A., & Calcaterra, C. (2017). Crypto transaction dispute resolution. The Business Lawyer, 73(1), 109-152.
Kasap, G. H. (2021). Can artificial intelligence ("AI") replace human arbitrators? Technological concerns and legal implications. Journal of Dispute Resolution, 2021(2), 209–254.
Khokhar, M. A., Anees, A., & Gujjar, M. W. (2025). From Judges to Algorithms: The Future of Corporate Adjudication in AI-Driven Courts. The Critical Review of Social Sciences Studies, 3(2), 1794-1811.
Kleros. (2026a, February 13). Kleros in the agentic society | AI jurors, kill switches & the future of dispute resolution [Video]. YouTube.
Kleros. (2026b, February 11). Proof of humanity.
Koulu, R. (2016). Blockchains and online dispute resolution: smart contracts as an alternative to enforcement. SCRIPTed, 13, 40.
Landau, T. (2003). The requirement of a written form for an arbitration agreement when "written" means oral. ICCA Congress Series (No. 11, p. 48).
Lehmann, M. (2025). The New York Convention's Borderline. Blockchain Arbitration and Artificial Intelligence. Transforming Arbitration: Exploring the Impact of AI, Blockchain, Metaverse and Web3, 71-89.
Lesaege, C., & Ast, F. (2019). Short Paper v1. 0.7 Clément Lesaege, Federico Ast, and William George September 2019.
Lowther, J. (2020). Recognition and enforcement of Kleros awards under the New York Convention in developing areas. (Kleros Fellowship of Justice, 3rd Cohort). Kleros.
Marenco, E. A. C., Castiblanco, M. L. P., & Suárez, J. S. C. (2025). Viabilidad de la implementación de cibertribunales en los procesos de arbitraje. Misión Jurídica, 18(28).
Martinez, A. L. (2025). Between Judge and Algorithm: Key Risks of Automated Decision-Making in the Digital Rule of Law. Available at SSRN 5749802.
Mohammed, M. A., De-Pablos-Heredero, C., & Montes Botella, J. L. (2024). A Systematic Literature Review on the Revolutionary Impact of Blockchain in Modern Business. Applied Sciences, 14(23), 11077.
Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system.
Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. SSRN.
Ohnoutková, B. (2025). LIMITATION ON THE USE OF ARTIFICIAL INTELLIGENCE (AI) IN ARBITRATION UNDER THE NEW YORK CONVENTION. The Lawyer Quarterly, 15(4).
Papadopoulos, N., Vasilaki, M., & Kotsias, Y. (2025). Operational AI in Business Excellence from Theory to Measurable Results. National Journal of Quality, Innovation, and Business Excellence (NJQIBE), 2(1), 44-54.
Pinheiro, L. D. L. (2025). Arbitration, Blockchain and Artificial Intelligence. Available at SSRN 5272787.
Prawira, C., & Lewiandy. (2025). Crowds, code, and courts: Evaluating the procedural fairness of Kleros as decentralized justice platform. Syiah Kuala Law Journal, 9(1), 392–403.
Qadri, T. (2025). THE ROLE OF ARTIFICIAL INTELLIGENCE SYSTEMS IN SMART ARBITRATION. Russian Law Journal, 13(1), 62-70.
Reddy, S., Müller, A., & Roberts, M. (2023). Platform Courts and Online Dispute Resolution: Redefining Access to Justice in the Digital Era. Legal Studies in Digital Age, 2(2), 49-60.
Santana Galindo, D. I. (2020). Decentralized Justice in International Arbitration (Kleros Fellowship of Justice: 3rd Cohort). Kleros.
Schmitz, A. J. (2026). Promise and Pitfalls of AI In Mass Arbitration. Ohio State Legal Studies Research Paper, (962).
Schmitz, A., & Rule, C. (2019). Online dispute resolution for smart contracts. J. Disp. Resol., 103.
Shih, S., & Chang, E. C. R. (2024). The Application of AI in Arbitration: How Far Away Are We from AI Arbitrators?. Contemp. Asia Arb. J., 17, 69.
Shoukat, D. (2025). Using AI in International Arbitration: From Predictive Analytics to Automated Awards. Legal Research & Analysis, 3(2), 5–12.
Swain, P. (2025). The artificial intelligence and machine learning blueprint: Foundations, frameworks, and real world applications. Deep Science Publishing.
Treacy, E. (2022). The effectiveness of artificial intelligence in simplification of arbitration proceedings: fiction or seventh seal in the world of arbitration?. Int'l JL Ethics Tech., 137.
Virues, M. (2021). Accommodating Kleros As A Decentralised Dispute Resolution Tool For Civil Justice Systems: Theoretical Model And Case Of Application. (Kleros Fellowship of Justice, 3rd Cohort). Kleros.
Volokh, E. (2019). Chief justice robots. Duke law journal, 68(6), 1135-1192.
Walters, R. (2025). Robots replacing human arbitrators: the legal dilemma. Information & Communications Technology Law, 34(2), 129-145.
Waste Management, Inc. c. Estados Unidos Mexicanos. (2000, 2 de junio). Laudo arbitral, CIADI, Caso No. Arb. (AF)/98/2, 21, p. 15.
Zhuk, A. (2023). Applying blockchain to the modern legal system: Kleros as a decentralised dispute resolution system. International Cybersecurity Law Review, 4(3), 351-364.
Zhuk, A. (2026). Enhancing online dispute resolution through natural language processing: A case study of Kleros (Kleros Fellowship of Justice, 8th Cohort). Kleros.