Prediction markets are experiencing remarkable growth. Yet their success depends on a crucial question: who decides whether a prediction has come true? Polymarket, a leading prediction market, relies on decentralized voting to resolve disputed outcomes. We argue that this mechanism suffers from fundamental due process deficiencies, particularly a lack of independence of decision-makers whose financial interests may conflict with impartial adjudication. We examine these deficiencies and consider two reforms: independent human adjudication and AI-based dispute resolution.
The Rise of Prediction Markets
Prediction markets are booming. Platforms such as Polymarket and Kalshi allow participants to trade contracts whose payouts depend on the outcomes of specified future events. These range from election results and geopolitical developments to economic indicators and corporate decisions. By aggregating participants’ expectations, market prices can provide information about the likelihood of these events. The commercial potential is attracting considerable investor interest. In July 2026, The Economist 1843 reported that Kalshi and Polymarket were seeking funding at valuations of approximately $40 billion and $15 billion, respectively.
But who determines whether the predicted event has actually occurred? And what happens when the answer is disputed? Prediction-market contracts may contain ambiguities that become apparent only when the market must be resolved. Consider the controversy surrounding a Polymarket contract asking whether Ukrainian President Volodymyr Zelenskyy would wear a suit before July 2025. What qualifies as a suit? When substantial amounts of money depend on the answer, participants have strong incentives to favor competing interpretations. Disagreements about the interpretation of market rules can turn seemingly simple questions into contentious disputes.
A credible dispute resolution mechanism is therefore essential to the further development of prediction markets. Luana Lopes Lara, co-founder of Kalshi, is quoted in The Economist’s 1843 as describing the platform’s ability to define and resolve bets clearly as ‘existential’. As Bloomberg reported in May 2026, nearly 2,000 Polymarket contracts had been disputed over the preceding year. In April alone, 230 contracts involving more than $1 billion in trading volume were resolved through UMA. The integrity of the resolution process is thus not merely a matter of procedural niceties. It is fundamental to the economic viability of prediction markets.
Polymarket’s Dispute Resolution System
Polymarket’s international, blockchain-based platform uses UMA’s Optimistic Oracle as its principal mechanism for resolving prediction-market contracts. A proposed outcome is submitted to the oracle, together with a financial bond. If nobody challenges the proposal within the prescribed period, it is accepted. If the proposal is challenged and the dispute proceeds to UMA’s Data Verification Mechanism (DVM), holders of UMA tokens vote to determine the outcome.
The system replaces a conventional adjudicator with a decentralized voting process. Voting power depends on the amount of UMA tokens staked. The mechanism seeks to incentivize truthful reporting by rewarding voters who support the prevailing outcome. It draws on Thomas Schelling’s concept of focal points, developed in The Strategy of Conflict (1960), which explains how individuals may converge on a common choice without communicating, by anticipating what others will regard as the natural solution.
The attraction is evident. Disputes can be resolved without appointing judges or arbitrators for each case. The system promises speed, scalability, and low costs. But does a mechanism designed to coordinate votes also provide procedural justice?
The Due Process Deficit
Procedural justice requires more than reaching a decision. Comparative civil procedure and international commercial arbitration recognize fundamental requirements of independence and impartiality, equal treatment, and the opportunity to be heard. Transparency and reasoned decisions provide further safeguards against arbitrary decision-making. These principles are reflected in the ALI/UNIDROIT Principles of Transnational Civil Procedure and the UNCITRAL Model Law on International Commercial Arbitration, for example. They do not necessarily apply directly to every private dispute resolution mechanism. Nevertheless, they provide an appropriate benchmark for assessing its procedural legitimacy.
Measured against these standards, Polymarket’s system suffers from three principal deficiencies.
First, the transparency of the decision-making process is limited. Although blockchain technology makes voting transactions publicly observable, it does not reveal who ultimately controls the voting wallets, whether voters have financial interests in the disputed outcome, or whether they coordinate their decisions. The public record of voting transactions therefore provides only a partial picture of how decisions are made.
Second, and more fundamentally, the decision-makers are not independent. UMA token holders have financial interests and may vote on disputes in which they hold positions in the underlying prediction market. Moreover, voting power is proportional to token holdings rather than distributed equally among decision-makers. Wealthy participants can therefore acquire considerable influence over dispute resolution.
The concentration of voting power compounds this problem. A Bloomberg investigation published in May 2026, based on blockchain data, identified nine cryptocurrency wallets exercising disproportionate influence over disputed Polymarket outcomes. Although wallets do not necessarily correspond to distinct individuals, and concentration alone does not establish manipulation, the ability of financially interested participants to exercise decisive influence raises fundamental concerns about impartiality and conflicts of interest.
The economic incentives create an additional problem. A rational voter may be more interested in anticipating the majority decision than in independently establishing whether the relevant event occurred. This recalls John Maynard Keynes’s famous beauty-contest analogy in The General Theory of Employment, Interest and Money (1936), where he describes investors seeking to anticipate what average opinion expects average opinion to be, rather than identifying fundamental value. A similar logic governs UMA voting.
The implications extend beyond the voting process. Prediction-market participants may trade on their expectations about how UMA voters will resolve a disputed contract, rather than solely on their assessment of the underlying event. Conversely, market prices may influence voters’ expectations about the likely outcome. The result could be a self-reinforcing feedback loop in which market prices influence voting behaviour and anticipated voting outcomes influence market prices.
This process could be particularly susceptible to the influence of large token holders. A token holder who can demonstrate control over substantial voting power could trigger a sudden shift in expectations by publicly announcing its preferred resolution, as other participants anticipate that voters will follow its lead. The resulting convergence of expectations could influence both market prices and the final resolution, regardless of the merits of the underlying dispute.
Third, the system lacks authoritative reasoned decisions. A voting outcome does not explain why a particular interpretation of the market rules prevailed. It does not reveal how competing evidence was evaluated or why alternative interpretations were rejected. Without reasons, it is impossible to assess the basis of individual decisions or establish whether comparable disputes have been resolved consistently. This also undermines accountability.
As we argued in an earlier OBLB post, guessing what others will decide is fundamentally different from adjudication. A process that rewards participants for anticipating the majority outcome may sometimes produce correct decisions. But correctness of individual outcomes does not establish the procedural legitimacy of the mechanism that produced them.
Of the deficiencies identified above, the lack of independence is particularly troubling because its consequences extend beyond procedural legitimacy. It potentially undermines the confidence on which prediction markets depend. Participants who believe that disputed outcomes can be manipulated by financially interested voters may become reluctant to trade. The problem could ultimately threaten the economic foundations of the platform itself.
Reforming the System
How could Polymarket improve its dispute resolution mechanism? We see two principal alternatives.
Back to Traditional Justice
The first would be to replace token-holder voting with independent human adjudication. Polymarket could establish a standing panel of qualified arbitrators, drawing on institutional models such as the Court of Arbitration for Sport (CAS) in Lausanne, which maintains a roster of specialist arbitrators and provides for expedited proceedings. Disputes would be allocated according to predetermined rules. Decision-makers would be subject to conflict-of-interest requirements and obliged to provide reasoned determinations. Parties would have an opportunity to present their positions and challenge relevant evidence.
A properly constituted arbitration mechanism could provide established due process guarantees and internationally enforceable awards under the New York Convention, provided the determinations qualify as arbitral awards under the applicable law. To accommodate the volume and speed of prediction-market disputes, procedures would have to be radically streamlined, with substantial AI assistance in evidence gathering, case analysis, and drafting, as discussed in other work. Even then, the costs and delays associated with human adjudication could remain substantial, particularly for disputes involving relatively small amounts. Moreover, conventional arbitration may be an awkward fit for disputes primarily concerned with determining whether specified events occurred. A contractual determination mechanism offers greater flexibility.
Forward to AI Justice
For these reasons, a more promising alternative may be to resolve disputed outcomes through a contractual AI oracle operating outside conventional arbitration. Polymarket could designate an AI-based oracle in its contractual terms and authorize it to determine the outcome of disputed markets. The oracle would interpret the contractual conditions, gather and evaluate relevant evidence, and determine whether those conditions have been satisfied. It could also be required to provide a reasoned determination, identifying the evidence and interpretations on which its conclusion rests. Unlike token-holder voting, the outcome would not depend on the distribution of governance tokens.
The potential advantages are considerable. AI systems can process large amounts of information rapidly and at relatively low marginal cost. Recent research suggests that AI could already perform some of the tasks involved in resolving prediction-market disputes. In a study published in April 2026, Junhao Wen and co-authors found that web-enabled large language models achieved 89.58% agreement with UMA’s final resolutions in the disputes examined.
This finding is encouraging, but it requires careful interpretation. Agreement with UMA does not establish that the AI reached the correct outcome. If UMA decisions are affected by the deficiencies identified above, reproducing those decisions may reproduce their errors. Moreover, AI systems remain vulnerable to inaccurate information, misleading sources, and deliberate manipulation. Where substantial financial interests are at stake, participants may have incentives to spread false information deliberately to influence the AI oracle’s determination.
Institutional design would therefore be crucial. Polymarket could specify in its contractual terms which AI system would resolve disputes, how it would operate, and under what conditions it could be modified. An independent organization could administer the system, ensuring that neither Polymarket nor market participants could interfere with individual decisions. To protect the oracle against manipulation, it could be required to draw on trustworthy sources, verify contested information against independent evidence, and disregard attempts to influence its decisions through fabricated reports or misleading instructions. Each determination would identify the evidence considered and explain the reasons for the outcome. The system’s operation and decisions would be documented and subject to independent scrutiny. These safeguards would be essential to establishing the trust of market participants, a central issue for AI-based contracting more generally, as discussed in other work. Their implementation would be demanding but appears technically and institutionally feasible.
Conclusion
Polymarket’s dispute resolution mechanism suffers from fundamental procedural deficiencies, particularly the lack of independence of decision-makers. A contractually agreed AI oracle offers a promising alternative. With appropriate safeguards to ensure independence, transparency, and protection against manipulation, it could provide fair, fast, and inexpensive dispute resolution without recourse to conventional arbitration. Prediction markets should enable participants to predict future events. They should not require participants to predict how financially interested decision-makers will vote on whether those events occurred.
Anna-Sophie Hochgürtel is a DPhil in Law student at the University of Oxford.
Horst Eidenmüller is Statutory Professor for Commercial Law at the University of Oxford and Professorial Fellow of St Hugh’s College, Oxford.
Share: