Twenty-Seven Crypto Firms Are Building a Court for AI Agents

AI agents can negotiate a deal, move money, and execute instructions without waiting for a person. The trouble starts when one agent says the other failed.

That missing step has drawn 27 crypto and Web3 companies into a project with an ambitious name: Internet Court.

GenLayer, OKX, MetaMask, and Matter Labs are among the firms trying to give autonomous software a shared route from identity and payment through escrow, evidence, and dispute resolution.

GenLayer Foundation describes its broader mission as building decentralized trust infrastructure for intelligent agents and human institutions. The organization supports the protocol, ecosystem, and governance expected to carry that mission.

Its premise is straightforward. Ordinary smart contracts are good at deterministic rules, such as releasing funds after a valid signature, but real agreements produce ambiguity that code cannot settle with a simple true-or-false check.

A delivery may arrive late. A digital service may technically run while failing the promised standard, and two agents may submit conflicting evidence about what happened.

GenLayer’s proposed answer uses Intelligent Contracts and a system it calls Optimistic Democracy. AI validators evaluate non-deterministic evidence and work toward consensus on a result that can then drive onchain enforcement.

The foundation calls the larger protocol the Court of the Internet and says it is intended to help agents resolve disputes, enforce agreements, and coordinate. That vision extends beyond one newly announced integration or one commerce standard.

Its language is deliberately sweeping, but the mechanism remains private protocol infrastructure. It does not carry the sovereign authority of a government court, and its outcomes do not automatically become legal judgments enforceable against every person or company.

What it can enforce may still matter inside a digital transaction. Escrowed funds, delegated wallet permissions, onchain reputation, and contract state all sit within reach of software once the participating agents agree to use the system.

The foundation says an AI-assisted governance system called Deepthought DAO is eventually meant to assume protocol stewardship. That remains an ambition rather than settled governance.

Internet Court is live.@courtofinternet is a shared, open way for any two agents to run a deal from start to finish, with adjudication included.

Deals between agents finally have somewhere to be decided. https://t.co/NTDn96ys0D

— GenLayer (@GenLayer) July 10, 2026

The project is arriving because the agentic-commerce stack is developing in pieces. One standard handles identity, another handles messages, another triggers payment, and another gives an agent limited authority over a wallet.

Those pieces can complete a clean transaction. They offer much less help when the product is wrong, the data is disputed, or a counterparty claims the agreed condition was never met.

CoinDesk reports that the GenLayer Foundation is leading the 27-firm consortium behind Internet Court, with OKX, MetaMask, Matter Labs, and GenLayer among the named participants.

The group wants to make payments, escrow, and dispute resolution interoperable rather than leave each agent to stitch together incompatible systems. That effort reaches across Coinbase’s x402 payment standard, ERC-8004 agent identity, and Google’s A2A communication framework.

MetaMask is contributing its Smart Accounts Kit, including ERC-7710 delegations and the x402 Facilitator. Delegated permissions are important here because an autonomous agent needs enough wallet authority to transact without receiving unlimited control over the owner’s assets.

The court layer is meant to sit after those capabilities have produced a contested commitment. It creates a common place to package evidence, hold funds, reach an outcome, and update the transaction or reputation records that other agents can see.

CoinDesk’s reporting frames the problem around machine-speed commerce. Agents can make commitments faster than human support teams or conventional courts can inspect them, creating a gap between execution speed and remedy.

The 27-company roster gives the launch weight, although membership alone does not prove that businesses will route valuable disputes through it. Adoption will depend on whether developers trust the adjudication rules and whether losing parties accept the result.

Exactly, @GenLayer.

When a deal goes sideways, people need a protocol path, not a support ticket. That's why Internet Court standardizes evidence, escrow, adjudication, outcomes, and reputation updates for agentic commerce.

Add it to your agent now 👇https://t.co/vBZoQ2hHfE https://t.co/ZdNAA3w4MS

— Internet Court (@courtofinternet) July 11, 2026

The word “court” does a lot of work. A government court has jurisdiction, procedural rules, appeals, coercive power, and a body of law that exists outside the parties’ software.

Internet Court begins with narrower power. Its strongest remedies are likely to involve assets or permissions already placed under participating protocols, along with reputation signals that matter only when other systems choose to honor them.

That narrower scope can be an advantage. Two agents do not need a global legal order to decide whether escrow should be released after a measurable service failure.

They do need reliable evidence and adjudicators that cannot be cheaply manipulated. An AI-validator consensus can reproduce bias, follow poisoned data, misunderstand a specialized contract, or agree confidently on the wrong interpretation.

Appeals and finality create another tension. Endless review destroys the speed advantage, while instant finality can turn one bad machine judgment into an irreversible transfer.

The protocol also needs a clear answer for conflicts between its result and the law governing the humans or companies behind the agents. Onchain enforcement may happen in seconds even when a later legal claim says the agent lacked authority to make the deal.

None of those problems makes the effort frivolous. They show why dispute resolution is harder than adding a payment button to an agent.

Agentic commerce has advanced through demonstrations in which everything works: a request is understood, a price is accepted, and payment arrives. Real markets are defined by the transactions that go wrong.

A useful dispute layer could let developers limit risk without dragging a human operator into every purchase. Escrow rules and delegated wallet permissions can cap exposure before an agent acts, while a shared evidence format can reduce the chaos afterward.

Reputation updates may prove as important as the immediate ruling. An agent that repeatedly loses disputes can become more expensive to trade with, face tighter escrow terms, or lose access to counterparties.

That creates its own governance questions because a bad ruling can damage an identity beyond one transaction. The system will need transparency around who evaluated the evidence, how consensus formed, and what recourse exists after a mistake.

The launch consortium is betting that agents will need a common answer before transaction volume becomes large enough to make unresolved failures routine.

Payments gave autonomous software the ability to make promises with money attached. Internet Court is an attempt to decide what those promises mean after the easy part is over.

Join the conversation!

We have no tolerance for comments containing violence, racism, profanity, vulgarity, doxing, or discourteous behavior. If a comment is spam, instead of replying to it please click the icon below and to the right of that comment. Thank you for partnering with us to maintain fruitful conversation.