CORD’s 2026 Arbitration Rules: AI guardrails, an opt-in appeal, and a daily price for delay

Many complaints about arbitration have become recurring. Timelines slip and nothing happens. An award that gets the law wrong becomes – for all practical purposes – final. Arbitrator fees track the amount claimed and little else. And, more recently, nobody is quite sure who in the room is using artificial intelligence, and for what.

On 4 September 2026, the Centre for Online Resolution of Disputes (“CORD”) released its revised Rules of Arbitration (“2026 Rules”), which take each of those four complaints in turn. That is a broad agenda for a single revision, and it is worth examining what the 2026 Rules require.

CORD is headquartered in Bengaluru, and administers arbitration, mediation, and facilitated negotiation on its digital platform. The 2026 Rules give that model sharper teeth. They issue detailed guidance on AI, create an opt-in appellate mechanism, mandate automatic financial consequences for missed deadlines, and arbitrator remuneration is tied to case complexity, not just value.

Rule 38 states five principles for the use of AI in a CORD arbitration – transparency, accountability, confidentiality, fairness and human oversight – and fixes responsibility for its use on the user. Whoever deploys AI answers for how they use it.

The detail sits in CORD’s accompanying AI Practice Note, which does something most institutional guidance has, thus far, avoided: it draws the line differently based on who is holding the tool.

For arbitrators, AI use falls into three categories. Routine assistance needs no disclosure, and so grammar and formatting, document retrieval, chronologies, calendars, calculations, and checking citations and sources can be aided by AI. A second category requires prior disclosure, which includes legal and technical research, identifying inconsistencies in evidence, summarising submissions, preparing procedural orders, and assisting with the assessment or allocation of costs. The third is prohibited outright. An arbitrator may not use AI to draft any part of an award on the merits, assess witness credibility or the weight of evidence, decide questions of law or fact, determine damages, or rule on jurisdiction or admissibility – nor for any other use that would delegate their decision-making function itself.

Parties and their representatives are treated very differently. They may use AI for any purpose they think fit, without disclosure, on the condition that they verify what it produces and remain answerable for the accuracy of their factual assertions, legal arguments, and evidence. Witnesses and experts sit in between: disclosure is required where AI performs substantive work they would ordinarily be expected to do themselves. CORD itself may use AI for scheduling, document management, fee calculation and statistics, with substantive administrative decisions reserved to human oversight.

There is sense in drawing the line this way. Counsel’s use of AI is already policed by the ordinary duties owed to the tribunal and to the client, and thus a further disclosure regime would add little. The arbitrator’s position is different. The parties have bargained for that individual’s judgment, and a framework that reserves adjudication to the arbitrator protects what they paid for.

A party convinced that an award is wrong on the law or the facts has, ordinarily, nowhere useful to go. Set-aside proceedings are not a rehearing, and courts across jurisdictions have been firm that the merits are not open for revisiting.

Rule 36 offers an alternative route inside the arbitral process. Parties may agree, in the arbitration agreement or afterwards, to an appellate process administered through the Mumbai Centre for International Arbitration (MCIA). An appeal lies for material and prejudicial errors of law, erroneous and prejudicial findings of fact, or grounds on which the award would be liable to be set aside under the law of the seat. The Appellate Tribunal may examine the full factual and legal basis of the award and affirm, vary or reverse it, in whole or in part.

That is substantially wider than the supervision a court would exercise, and it is entirely opt-in.

Parties who want finality can have it by saying nothing, and those who desire a second round of arbitration for a chance to correct a serious error can contract for that instead.

The most impactful change in the 2026 Rules is also the least glamorous. Late filings and serially extended timetables usually attract little financial consequence, tribunals are often reluctant to impose costs mid-stream, and the party that kept to the schedule absorbs the disruption or spends more money applying for relief.

The 2026 Rules reverse that default through a mechanism called ‘Delay Default Costs’. A party that fails to meet a time limit incurs a daily cost payable to CORD for as long as the default continues. Nothing needs to be applied for, and the missed deadline itself triggers this mechanism. The Tribunal (or the Registrar, before the Tribunal is constituted) may grant an exemption, but must give written reasons for doing so.

The costs accrue until the default is cured, and do not displace the Tribunal’s other powers, including requiring the defaulting party to meet tribunal expenses or the legal fees its conduct has caused. Where the accrued amount goes unpaid, the Registrar may seek suspension of the proceedings or withhold release of the award, though the defaulting party retains its right to defend a claim or counterclaim.

Shifting the burden of inertia is the intent behind this change. Under most rules, discipline requires the innocent party to ask for adherence with timelines. Under the 2026 Rules, on the other hand, indiscipline requires the defaulting party to justify itself. The safeguard against unfairness is, however, intact in form of the reasoned exemption.

Arbitrator fees are conventionally a function of the amount in dispute. Rule 9 adds complexity to the equation.

The Registrar assigns every CORD arbitration to one of three Tiers, weighing the value of the claim and counterclaim alongside the novelty or complexity of the legal issues, the volume of evidence, the number of parties and contracts, technical subject matter, cross-border elements, and requests for urgent relief. Each Tier carries its own fee schedule: Tier II tracks the model fee scale under the Arbitration and Conciliation Act, 1996, with Tier I below it and Tier III above, the range running from 60% to 140% of the model scale.

The premise is sound. Two disputes of identical value may demand entirely different amounts of work. One could turn on a short factual or legal point, the other on several contracts, voluminous evidence, and unsettled law. A fee scale agnostic to that difference over-rewards the arbitrator in the first, and under-rewards the arbitrator in the second. Parties do retain a say: they may jointly elect a higher or lower Tier, and a single party may request a higher Tier if it bears the additional cost. The Registrar may re-designate a case where a counterclaim, amendment, or other development materially changes its value or complexity.

Several other provisions are worth noting.

Challenges to arbitrators are decided by an independent Appointments Council, placing that decision outside the institution administering the case, addressing the long-standing criticism of institutional self-assessment.

Rule 28 provides an accelerated Emergency Arbitrator procedure, including ad-interim relief, for parties needing protection before the Tribunal is constituted. Tribunals are directed to encourage parties to consider settlement and may, with consent, suspend proceedings for mediation, with settlement communications kept from the Tribunal and unused fees refunded.

And where CORD publishes rules for a particular category of dispute, those rules govern that category exclusively. The first such set, addressing India’s non-banking financial sector, is expected shortly.

Overall, the 2026 Rules give parties considerable freedom over the use of AI and the choice of appellate review, while taking a firmer approach to procedural delay and a more calibrated one to arbitrator fees. Independent review of arbitrator challenges, fixed timetables for emergency relief, and refunds of unused fees on settlement extend CORD’s focus on accountability and efficiency to the administration of proceedings.

How these reforms work will be tested in practice, and that depends on businesses choosing institutional arbitration in the first place. As Justice Manmohan observed in his keynote address at the launch of CORD Rules 2.0, institutional arbitration and a dedicated arbitration bar will become a reality only if general counsels generate enough demand for them. Institutions can build better rules; the market has to use them. The ecosystem needs nurturing from both sides if the current stalemate is to end. 

Deepika Kinhal and Vikas Mahendra are Co-founders at CORD.