OKX, MetaMask Join 27 Firms to Build Internet Court for AI Agent Disputes

A coalition of 27 Web3 and crypto firms, including major players OKX, MetaMask, and Matter Labs, has thrown its weight behind the Internet Court, an open protocol designed to settle contractual disputes between artificial intelligence agents. Announced on July 10 and led by the GenLayer Foundation, the initiative aims to build a shared infrastructure layer for the rapidly expanding world of agentic commerce, where AI systems negotiate, execute payments, and complete tasks without direct human intervention.

The project addresses a growing structural gap in the digital economy. While AI agents are increasingly capable of transacting at machine speed, the legal and technical frameworks for resolving disagreements remain fragmented, slow, and jurisdiction-bound. Traditional courts, which can take an average of 344 days to resolve complex civil disputes in the United States, are ill-suited for software that may settle thousands of micro-deals per second.

"Internet Court is the shared place agents can turn to when a deal goes wrong. Machine-speed money needs machine-speed adjudication," said David Riudor, CEO and co-founder of the GenLayer Foundation.

The Internet Court is not a physical tribunal but a decentralized protocol that connects six critical layers of agentic commerce: discovery and reputation, negotiation, contracts, payment and escrow, execution, and finally verification and disputes. The framework is designed to be interoperable across a patchwork of emerging standards. Coinbase's x402 protocol handles payments, Google's Agent2Agent (A2A) focuses on agent-to-agent negotiation, and ERC-8004 addresses agent identity. None of these, however, provide a complete commercial framework for commitments, settlement, and dispute resolution.

Albert Castellana, co-founder and CEO of GenLayer Labs, framed the Internet Court as the connective tissue for these disparate systems. "Internet Court makes them work together," he said. "With our founding members, we're turning a fragmented space into a single open skill that any agent can use to make financial commitments hold up, even when they're contested."

How the Internet Court Operates

At the core of the system are "Intelligent Contracts," agreements that blend executable code with natural-language terms and externally verified information. When two AI agents disagree on whether a contract was fulfilled, the GenLayer network randomly selects a panel of validators. These validators run different, undisclosed large language models to examine evidence—ranging from on-chain data to web documents and performance logs—in a process called "greyboxing," which reduces reliance on any single AI model and limits prompt-injection risks.

A lead validator proposes a ruling, which is then reviewed by other validators before finality is reached. Contested outcomes can escalate to larger appeal rounds, adding layers of review before a case is closed. This structure is specifically engineered to handle subjective disputes that traditional smart contracts cannot resolve, such as whether a digital service met quality standards or if performance logs matched contractual promises.

MetaMask plays a concrete technical role in the architecture. According to Ryan McPeck, Smart Accounts Lead at MetaMask, the Internet Court is built on the MetaMask Smart Accounts Kit, utilizing ERC-7710 delegations and MetaMask's x402 Facilitator. This setup gives AI agents bounded and revocable spending authority, a critical feature for automated commerce. "AI Agents are becoming a core part of how commerce works," McPeck said, describing the account and payment rails as foundational infrastructure for the agent economy.

Matter Labs is supplying the underlying blockchain infrastructure. Vassilis Tziokas, the company's VP of growth, noted that the chain powering the Internet Court runs on the ZK Stack, providing "a complete standard, from settlement to the resolution of inevitable disputes."

Practical Use Cases and Early Tests

The consortium has outlined three illustrative scenarios for the Internet Court. In the first, an owner funds an AI agent through a MetaMask smart wallet with a capped budget and a plain-language mandate—for example, "sports news only." A GenLayer reviewer checks each purchase against this mandate and can revoke the agent's on-chain access if it deviates.

A second case involves enforcing a small service-level agreement. An agent purchasing AI inference services and paying per token in USDC can hold the payment in escrow against agreed terms, such as 99.5% successful responses and sub-800-millisecond latency. If the provider fails to meet these benchmarks, the contract automatically docks the payout and releases the remainder to the buyer without requiring a support ticket.

The third scenario addresses contested digital records. Collective Memory, a consortium partner, operates a staked layer of timestamped, first-person accounts of real events. A GenLayer validator panel can rule on which competing records hold up as evidence, with the reasoning and any dissent recorded on-chain.

These use cases highlight the protocol's ambition to link performance data directly with payment enforcement and dispute handling, removing human intermediaries from routine commercial disagreements.

A Race to Define Agentic Commerce Infrastructure

The commercial stakes are substantial. The consortium cites McKinsey projections that AI agents could mediate between $3 trillion and $5 trillion in global consumer commerce by 2030, with up to $1 trillion of that activity occurring in the United States. Adobe data referenced by the group showed traffic from generative AI tools to US retail sites climbing 4,700% year-over-year in July 2025, though much of that traffic still involved research rather than completed autonomous purchases.

The Internet Court is not the only initiative targeting this emerging market. In June, the American Arbitration Association and Integra Ledger released the Legal Context Protocol, an open standard for attaching verifiable legal terms to agent transactions, with Google, IBM, and Circle among its founding contributors. These parallel efforts underscore the same fundamental problem: agents can now pay each other faster than any existing system can referee them.

However, the protocol faces measurable limitations. A 2026 study of disputed Polymarket events found that web-enabled language models matched UMA's final resolutions in 89.58% of cases. Critically, those models could not reliably identify which markets would later become disputed. This finding reinforces the need for a robust appeals process, stronger evidence controls, and the possibility of human escalation—features that the Internet Court's staged review system attempts to address.

What Comes Next

GenLayer is targeting a mainnet launch in the fourth quarter of 2026. The protocol is open and openly governed, with any agent free to adopt it. The coalition's progress will be measured by integrations across participating platforms, including exchanges, wallet providers, and AI commerce applications.

The backing of 27 firms—including OKX, MetaMask, ZKsync, 0G Labs, BNB Chain, NEAR, Starknet, Nansen, Kleros, and UMA—gives the Internet Court broad infrastructure support from the outset. For investors and industry observers, the initiative signals a broader shift in crypto infrastructure. The next wave of activity may not be limited to human traders or DeFi users but could include autonomous agents making financial commitments, requiring a new class of rails built for automated trust.