The situation concerning both “old” fossil fuels and “new” critical minerals as a result of the dual surge of disputes is nothing less than the traditional investment protection standards being put to unbearable and fundamentally contradictory stress. The Fair and Equitable Treatment (FET) standard, an ‘absolute’ and ‘non-contingent’ standard, is the most prominent and heavily litigated norm in international investment law.

The crux of the legal conflict has always been the investor’s wish for a stable regulatory environment versus the state’s indefeasible, sovereign “right to regulate” for the public good. It is a conflict that is now far from being just a hypothetical academic debate; it has become the principal issue in the 2025 wave of resource arbitrations and has caused what can only be termed as a “doctrinal whipsaw” for the FET standard. In “Stress Test 1,” fossil fuel investors are leading the way, and their argument is that by previously permitting fossil fuel investments, the states have created “legitimate expectations” of stability. They claim that such new climate measures, like coal phase-outs, violate these expectations and thus constitute a breach of the FET standard. In this scenario, FET is utilized to punish the states for their regulatory actions. At the same time, in “Stress Test 2,” investors in critical minerals assert that their licenses provided “legitimate expectations” of safety. They argue that state “resource nationalism” measures like the cancellation of licenses or, in some instances, the failure to clear protest blockades, violate these expectations and thereby infringe the FET standard. Here, investors exploit FET to penalize the states for their lack of regulation (that is, for not wielding state power to safeguard the investment). The state is thus exposed to liability for both action and inaction.

This scenario uncovers the fact that the “right to regulate,” despite being recognized by a great number of modern treaties as well as by legal scholars, is nothing more than an illusion in the absence of a “right to regulate affordably.” The chilling effect is not a result of the lack of the right to regulate, rather it stems from the potentially high cost that it might take to exercise the right. The presence of the nebulous notion of “legitimate expectations” ties the whole system now. Was an investor in 1995 legitimately expecting that the host country’s climate policy was going to stay the same for 30 years? Was an investor in 2018 justified in thinking that a country was never going to use its sovereign right to reclassify a strategic mineral? The wave of arbitrations in 2025 is a gigantic, investor-backed gamble that tribunals will keep interpreting “expectations” widely, thus providing the investment with more protection than the state’s right to govern.

Again, relying on Fellas, as cited in ONGC, Dhanuka invoked direct benefits estoppel, which prohibits a party from taking inconsistent positions or seeking to have it both ways by relying on the contract when it works to its advantage and ignoring it when it works to its disadvantage. 

i. Does Direct Benefits Estoppel Survive Cox & Kings?

Now, unlike the intertwined estoppel theory, it can certainly be used by a signatory against a non-signatory. However, the Supreme Court in Cox & Kings Ltd. v. SAP India, does not appear to favour this theory in at least two contexts. Firstly, under the group of companies’ context, the Supreme Court observed that “even though a subsidiary derives interests or benefits from a contract… they would not be covered… merely on the basis that it shares a legal or commercial relationship.” Secondly, under the broader umbrella of impleadment of third parties under “claiming through or under him” i.e., Section 8 of the A&C Act. The Supreme Court while examining Rinehart v. Hancock Prospecting Pty Ltd, where the Australian High Court adopted the estoppel approach (“a non-signatory party who elects to take the benefit of some aspects of the contract, must also accept the burden of it”),  held that this approach cannot be adopted “in the context of the phrase “claiming through or under” as doing so would be contrary to the common law position and the legislative intent underpinning the Arbitration Act”. Further, the Court clarified “claiming through or under” applies only to parties in “derivative capacity” (assignees or successors), not those merely deriving benefits.

Thus, two things are clear, the principle of benefits estoppel does not fit within the: (i) group of companies’ context (as benefit-derivation alone is insufficient); and (ii) “claiming through or under”, as it’s limited to entities in derivative capacity. Where direct benefits estoppel fits, if anywhere, remains unclear. Whether it operates independently, within group of companies’ contexts, or beyond both, Cox & Kings left unanswered.

ii. Direct or Indirect benefits?

But even assuming that direct benefits estoppel survived Cox & Kings, its application in Dhanuka must still satisfy the principle’s thresholds. Given it’s not provided in Indian jurisprudence, it is useful to look to U.S. jurisprudence, where this doctrine originates. The case of Life Techs. Corp. v. AB Sciex held that the principle was confined to direct benefits, and does not extend to indirect benefits. While distinguishing direct from indirect benefits it held: first, benefits are “direct” only when they “arise from the unsigned contract containing the arbitration clause”, and “indirect” “when merely incidental” to its performance; second, benefits are direct only “when specifically contemplated by the relevant parties”, and indirect when the non-signatory’s benefit was not within their original contemplation.

Dhanuka seems to not fulfil either step. The developer’s revenue-sharing and control mechanisms arose from the developer-maintenance agency agreement, not from the flat owner-maintenance agency agreements containing the arbitration clause. The Developer derives supervisory rights and profit-sharing from its contract with the Maintenance Agency, not from the Agency’s separate agreements with flat owners. These are indirect benefits incidental to the contractual relationship between two other parties.

Moreover, nothing suggests flat owners contemplated the developer benefiting from their maintenance agreements when executing them years after the agreement. The parties to the arbitration clause never manifested intent to bind the developer, absent from their contracts entirely.

The likely benefits of integrating AI in Indian arbitration should be evaluated based on its impact on efficiency and fairness in arbitral decision-making. Recent empirical and institutional developments suggest that AI can help bring efficiency to the arbitral process. However, if unregulated AI is used in arbitration to make decisions, it raises various concerns regarding transparency, bias, responsibility, clarity, and confidentiality. These concerns have a direct impact on the validity of the arbitral award.

Internationally, at the institutional level, we observe a careful yet inconsistent approach to the use of AI in arbitration. AI can be used as a supportive tool with human oversight and disclosure, according to the guidelines released by professional organisations like CIArb, SVAMC, and AAA-ICDR. However, some important institutions, such as the ICC, ICSID, LCIA, and SIAC, have not established rules to guide the use of AI in the arbitral process. This uneven approach points toward a general acceptance that although AI helps in procedural aspects, its role in core decision-making is debatable. 

Empirical studies show the difference between substitutive and assistive AI use. In the 2025 International Arbitration Survey by White & Case, 2,402 questionnaire responses and 117 interviews were collected from a diverse cross-section of the international arbitration community. Participants included in-house counsel from the public and private sectors, arbitrators, private practitioners, representatives of arbitral institutions, academics, tribunal secretaries, experts, and third-party funders. There is strong support for the use of AI in administrative tasks. 77% of respondents are in favour of utilising AI to determine interest, costs, and damages. 66% of respondents support using it to summarize submissions. It is important to note that only 23% of respondents are in support of using AI for legal reasoning, while the majority oppose using it to evaluate merits or credibility. This skepticism is backed by recent research showing that AI judges apply the law consistently and strictly, whereas human adjudicators consider a wider context and use moral reasoning in their decisions. The main concern is that AI biases and mistakes can go unnoticed, which is further worsened by the black box nature of large language models.

Apart from concerns about bias and explainability, using AI in arbitration presents serious challenges to the confidentiality and privacy that are essential to arbitral proceedings under Section 42A. Arbitration often involves sharing sensitive commercial information, trade secrets, and personal data. This makes deploying AI particularly delicate, especially when using third-party or cloud-based tools. The 2023 BCLP Annual Arbitration Survey highlights data protection and confidentiality as major concerns for arbitration users regarding AI adoption. In response, the SVAMC Guidelines stress that AI must be used in a way that respects confidentiality obligations. They also warn against processing confidential information without permission and proper safeguards. Therefore, confidentiality should be a key limit on AI-assisted arbitration, necessitating clear disclosure requirements, security standards, and restrictions on data retention to maintain the legitimacy of arbitration.

It is important to evaluate these concerns in the real world. Some cases show that blind trust in AI harms procedural integrity. In Mata v. Avianca, a US court sanctioned lawyers for using AI-generated fake citations. This case demonstrates problems of error and loss of trust when outputs are not verified. In LaPaglia v. Valve Corp., one of the parties challenged the award, alleging that the arbitrator used AI for reasoning, which is an unauthorized delegation of authority. Although these are not Indian cases, they point out that fairness, party autonomy, and independent thinking can be jeopardized by AI. These principles are highly valued by the Indian court for maintaining the legitimacy of the award.

Indian judiciary and policy discussions have taken a careful approach to AI. The Kerala High Court guidelines prohibit the use of AI in judicial reasoning because of data security, privacy, and public confidence. The Supreme Court of India’s Centre for Research and Planning, in its white paper on AI and the judiciary, advocates a governance framework centered on human-in-the-loop oversight, mandatory verification protocols, and transparency obligations whenever AI assistance is used. This cautious position was judicially reaffirmed in Kartikeya Rawal v. Union of India, where, while dismissing a PIL seeking regulation of AI in the judiciary, the Supreme Court categorically assured that AI would not be permitted to overtake judicial decision-making, emphasising that technology must remain strictly subordinate to human judgment. While these are judicial guidelines, they offer insights for the use of AI in arbitration since arbitral awards are reviewed under section 34 of the A&C Act. 

Unregulated use of AI in arbitration goes against the Indian arbitration law. Sections 18 and 31 mandate impartial treatment of parties and reasoned awards. AI systems that primarily rely on probabilistic pattern matching instead of true reasoning go against the mandate. Apple’s The Illusion of Thinking shows that a large reasoning model can mirror taught patterns but face problems with novel or complex situations. These limitations of AI make it difficult for an arbitrator to explain, defend, and accept accountability for decisions, thereby undermining transparency, accountability, and clarity.

This is not to completely negate the role of AI in Indian arbitration. The T.K. Viswanathan Committee treats AI as a helpful tool to reduce delay and procedural issues. The Pyrrho Investments Ltd. v. MWB Property Ltd. case shows the court’s acceptance of AI for technical tasks like predictive coding with human oversight. However, extending AI into substantive legal reasoning risks diluting statutory mandates under Sections 18 and 31 of the A&C Act, which require impartial treatment, intelligible reasoning, and demonstrable application of mind. Accordingly, AI assistance can be normatively justified only where it operates in an assistive, human-in-the-loop capacity, supported by standards on disclosure, verification, explainability, and accountability. Such a calibrated approach preserves efficiency gains while remaining faithful to the foundational principles of arbitral legitimacy under Indian law.

The reconsideration of Bharat Drilling is significant because prohibitory and “no-claim” clauses are a standard feature of government and public-works contracts. These clauses are intended to allocate risk ex ante and to limit exposure to specific heads of loss such as idle labour, idle machinery, or loss of profit. The widespread reliance on Bharat Drilling by tribunals and High Courts to dilute or bypass such clauses has created uncertainty and undermined contractual predictability in public procurement disputes.

At a doctrinal level, the reference reinforces the centrality of party autonomy. If parties have consciously agreed to exclude certain claims, allowing tribunals to disregard those exclusions risks rewriting the contract under the guise of arbitral discretion. The Supreme Court  observation that contractual limits “crystallise informed choices of parties” signals a clear concern that Bharat Drilling has been used to erode the sanctity of contract, contrary to the modern arbitration framework.

Until the larger bench settles the issue, Indian Builders serves as a caution against treating Bharat Drilling as a blanket authority and shows the need for tribunals and courts to engage closely with the language of the contract, giving due weight to party autonomy and the risk allocation expressly agreed between the parties.

Though the practice of centralised appointment system is doctrinally correct, yet it poses significant institutional and procedural concerns. Treatment of improper appointments as an incurable jurisdictional defect poses a significant risk of invalidating arbitral awards after parties have spent considerable time and resources. Furthermore, the centralisation of the appointment process with the SC creates a fragmented supervisory approach to arbitrations whereby various stages of an arbitration can be subjected to the supervision of different courts. For example, an application for interim relief under Section 9 of the Act may be sought before a jurisdictional High Court, arbitrator appointment before the SC under Section 11, and post-award challenges again before the court exercising jurisdiction under Section 34. Such fragmentation introduces outcomes, such as procedural complexity, institutional strain, and uncertainty for parties acting in good faith, which sit uneasily with the Act’s objectives of efficiency and predictability. It is these limitations that form the immediate backdrop to the proposed legislative intervention.