AI agents are beginning to negotiate agreements, arrange payments and execute transactions, but determining whether contractual obligations have been fulfilled presents a different challenge. A customer may commission work through an autonomous agent and subsequently dispute whether the delivered service meets the agreed standard. Both parties may have acted in good faith, yet deciding whether payment should be released still requires an assessment of the contractual terms and supporting evidence.

"Most of the contracts in the real world are ambiguous," said Albert Castellana, co-founder and chief executive of GenLayer Labs. He argues that conventional smart contracts, which execute predefined instructions automatically, are less suited to interpreting obligations that require judgement. GenLayer is developing what it calls intelligent contracts, combining executable code, natural-language conditions and external information to assess disputed outcomes. Its Internet Court initiative seeks to connect this capability with identity, payment and other contractual services. Both remain early-stage approaches rather than established institutional dispute-resolution arrangements.

Agent-initiated payments are already being tested within established financial infrastructure. In March 2026, Mastercard reported an authenticated transaction involving DBS and UOB in Singapore, in which an AI agent booked and paid for an airport transfer. The development illustrates the progress of automated payments, while disagreements over contractual performance require separate mechanisms for assessing evidence and determining outcomes.

Intelligent contracts seek to interpret obligations beyond fixed rules

Conventional smart contracts operate through deterministic instructions, producing outcomes based on predefined conditions and inputs. They can release funds when a specified requirement is met, but interpreting events outside a blockchain generally requires external data services, known as oracles. Even with access to that information, determining whether work meets an agreed standard may involve judgement that cannot be expressed through fixed rules.

GenLayer's intelligent contracts are designed to combine executable instructions with contractual terms written in natural language and information retrieved from publicly accessible sources. Castellana described how a contract could examine delivered work, compare it with agreed guidelines and assess whether the conditions for payment had been satisfied. The assessment could then trigger an action such as releasing funds held in escrow, where money is retained until specified conditions are met.

Castellana described Antseed, a project exploring whether signed operational records could be used to assess compliance with service-level agreements governing AI services. Where a customer disputes whether agreed performance standards were met, GenLayer's validators could examine the records and determine whether compensation should be released. The application remains under development rather than constituting an independently established commercial deployment.

The assessment depends on how the agreement defines acceptable evidence. Castellana explained that the parties, or their agents, must establish the contractual conditions and identify the information sources that can be consulted. GenLayer's protocol is intended to evaluate evidence within those parameters rather than determine the obligations independently. The reliability of an outcome therefore depends partly on the clarity of the original agreement and the accuracy of the information used.

AI validators assess disputed outcomes through consensus and appeals

Rather than relying on a single AI model, GenLayer is developing a network of validators, participants that use AI models to assess contractual evidence and vote on proposed outcomes. Castellana described an initial group of five validators, with one selected to examine the evidence and propose a decision. The others independently assess the proposal and vote on whether to accept it. Disagreements can trigger further evaluation, while appeals can involve additional validators reviewing contested outcomes.

GenLayer calls this process Optimistic Democracy, a consensus mechanism that combines independent assessments with procedures for challenging decisions before they become final. Castellana also described financial incentives intended to discourage validators from taking positions that diverge from the network's eventual decision. These mechanisms form part of the protocol's developing approach to automated adjudication, rather than an independently proven system for resolving contractual disputes.

Castellana argues that using different AI models reduces dependence on one model's interpretation. Agreement among validators, however, does not necessarily establish that a decision is correct. Models may rely on the same inaccurate information, share biases or interpret ambiguous requirements similarly. The reliability of an outcome depends on the evidence, the capabilities and independence of the models and the effectiveness of the review procedures. GenLayer has yet to demonstrate consistent results across a broad range of contractual disputes.

The availability of evidence also limits potential applications. Castellana acknowledged that GenLayer is initially concentrating on disputes involving publicly accessible information, while contracts requiring private or commercially sensitive records remain a longer-term development area. This matters where the assessment of contractual performance depends on confidential customer information or internal operational data.

Internet Court seeks to connect fragmented contractual services

Adjudication, the process of assessing a dispute and reaching a decision, addresses only one part of an automated agreement. The parties must also establish their identities, negotiate terms, arrange payment, determine how funds will be held and agree on procedures for handling disagreements. Castellana described these functions as fragmented across organisations developing separate protocols and services.

Internet Court, announced in July 2026, is an early-stage consortium initiative intended to connect these components through a common framework for transactions between AI agents. Participants include blockchain networks and dispute-resolution providers such as Kleros and UMA, which offer different mechanisms for reviewing contested outcomes. The initiative is exploring how agents could identify and combine these services rather than requiring every function to operate within a single protocol.

Castellana described a possible arrangement in which an agent establishes a contract on one blockchain, uses another service for payment and designates GenLayer as an initial mechanism for resolving disagreements, with escalation to another provider if necessary. The consortium is developing common specifications that would allow agents to select contractual and dispute-resolution services when creating agreements. The extent to which participating systems will operate together in routine commercial transactions remains to be established.

Related work is taking place through established dispute-resolution organisations. In June 2026, the American Arbitration Association (AAA) and industry partners introduced the Legal Context Protocol, an open standard intended to make legal terms, consent and dispute-resolution arrangements discoverable and verifiable when AI agents transact. The initiative addresses how agents identify contractual obligations and available recourse, while GenLayer is developing mechanisms for assessing disputed performance.

Automated decisions remain subject to legal and operational limits

Castellana argues that automated adjudication could reduce the cost of resolving relatively small contractual disputes. He estimated that straightforward assessments might cost less than a dollar, while extensive appeals could cost approximately $100 and take several hours. These are indicative estimates rather than independently verified operating results, and they do not establish a direct comparison with litigation or arbitration, which involve different procedures, protections and legal consequences.

"This is not trying to bypass the legal system," Castellana said. He envisages automated adjudication as a possible initial stage in resolving disagreements, with dissatisfied parties able to seek further review through other protocols, mediation or formal legal proceedings where available. A protocol may determine how funds held within an automated agreement are distributed, but that decision does not necessarily have the authority of a court judgement or legally enforceable arbitral award.

Established arbitration organisations are also experimenting with AI-assisted procedures. The AAA's AI Arbitrator examines case materials and prepares proposed findings and draft awards for human review, with an arbitrator retaining responsibility for the final decision. GenLayer is exploring a different model in which AI-validator consensus determines an executable contractual outcome, subject to its appeal procedures. The distinction concerns the allocation of decision-making authority, the procedures available to challenge outcomes and their legal recognition.

Castellana acknowledged that the regulatory treatment of automated adjudication remains uncertain and that human intervention and legal liability would continue to have a role. The ability to execute a decision through software does not resolve questions about the legal validity of contractual terms, responsibility for erroneous outcomes or the remedies available when a dispute extends beyond the conditions controlled by a protocol.

Contractual accountability will shape the next stage of development

GenLayer's approach seeks to bring the assessment of contractual performance into automated transactions, while Internet Court is exploring how dispute resolution can connect with the services through which agents establish agreements and transfer funds. The immediate challenge is whether these arrangements can produce consistent decisions when evidence is incomplete, contractual terms are disputed or AI models reach different conclusions.

As autonomous transactions extend across services and jurisdictions, their contractual frameworks will need procedures for reviewing decisions and assigning responsibility when outcomes are contested. The relationship between automated enforcement and recognised legal remedies will become increasingly important, particularly where disputes involve obligations or consequences beyond the funds controlled by a protocol.