Hook
A press release crossed my desk this week: the “Internet Court” has launched on Starknet. A protocol designed to arbitrate disputes between autonomous AI agents. The narrative is seductive. Agentic commerce—AI trading, negotiating, even suing each other—needs a judiciary. Starknet, with its native account abstraction and ZK-rollup efficiency, becomes the natural venue. But as someone who spent 2017 auditing ICO smart contracts and later designing governance frameworks for Aave, I have learned one thing: code that promises to resolve power struggles without revealing its own architecture is not a solution; it is a political statement in disguise.
We didn’t need another layer of abstract justice. We needed verifiable, auditable, and economically sustainable dispute resolution. This announcement provides none of that. It is a press release dressed as a protocol.
Context
The Internet Court, as described, is an application-layer protocol built on Starknet. Its target: the emerging ecosystem of autonomous AI agents that will conduct business—buying compute, renting storage, executing trades—on behalf of humans or other algorithms. When two agents disagree over a contract term, the court steps in. The decision, presumably, is enforced by on-chain asset transfers.
Starknet was chosen for two reasons: its ZK-rollup architecture offers high throughput and low fees, suitable for high-frequency agent transactions, and its native account abstraction allows AI agents to operate as first-class smart contract accounts, signing transactions without human intervention. The court itself is a set of smart contracts—though no code has been released.
The court claims to be a “structural milestone” for autonomous commerce. But structurally, what does it actually do? The answer is almost nothing, because the details are missing.
Core Insight: The Architecture of Missing Details
Let me conduct a forensic scan of what the announcement – and more importantly, what it omits – reveals.
First, innovation vs. substance. The concept of on-chain arbitration is not new. Kleros has been running decentralized jury systems for years. The Internet Court’s claimed innovation is specialization in AI agent disputes. But how? It does not specify how an “AI judge” or autonomous arbitrator would be trained, how evidence is submitted (on-chain or off-chain with oracles?), or how the judgment is enforced if the losing agent has already moved assets. Without these details, the innovation is purely nominal.
Second, the security assumptions. The court inherits Starknet’s security model, which rests on ZK-proofs aggregated on Ethereum. That is strong for settlement but says nothing about the correctness of the arbitration logic itself. If a bug in the voting contract allows a single agent to collude with the operator, the entire court becomes a rubber stamp for rent-seeking. I have seen this pattern before. In 2020, during the DeFi summer, a “decentralized” lending protocol had a governance exploit because the quorum threshold was coded as a fixed number rather than a percentage. The Internet Court could suffer from similar governance design flaws. We don’t know.
Third, tokenomics and incentives. The announcement is silent on any native token. If the court operates solely on STRK gas fees, who pays the arbitrators? Who is incentivized to submit evidence? Who decides which cases are frivolous? Without economic alignment, the system relies on altruism or foundation subsidies. Neither is sustainable. In my experience designing Aave’s quadratic voting mechanism, we learned that governance without skin in the game leads to apathy or capture. The Internet Court faces the exact same dilemma.
Fourth, the legality of the name. “Internet Court” is a brazen term. It implies jurisdictional authority. In the real world, courts are backed by state power. A smart contract can only enforce its rulings if assets are locked in escrow. What happens when an AI agent controlled by a Bermuda corporation refuses to pay? The “court” has no subpoena power, no contempt charges. The name is marketing, not a legal reality. This risk may seem minor, but regulators have recently targeted projects that overstate their legal status. SEC actions against “decentralized” platforms have often focused on misleading terminology.
Fifth, the missing team. The announcement names no founder, no development firm, no advisors. It could be a Starknet foundation-backed initiative, a solo developer’s side project, or a prank. The only entity visible is Starknet itself, which benefits from the narrative boost. As a DAO governance architect, I know that anonymity is acceptable for experimental protocols, but a dispute resolution system requires accountability. If the arbiter is anonymous, how do you appeal a bad ruling?
Sixth, the AI agent dependency. The entire value proposition rests on the existence of a thriving AI agent economy. That economy is not yet here. Most “agents” today are simple bots executing predefined tasks. We are years away from autonomous commercial entities that generate enough disputes to justify a dedicated court. The Internet Court solves a problem that may never exist at scale, while ignoring the immediate disputes that do exist: protocol governance disputes, oracle manipulation, cross-chain bridging faults. It is a solution looking for a problem.
Seventh, the competitive landscape. If the Internet Court gains traction, it will compete with Kleros (Ethereum-based), Aragon’s dispute resolution, and emerging AI-native arbitration protocols on other L2s like Arbitrum and Optimism. Starknet’s ZK advantages are real, but network effects matter. Arbitrum has deeper liquidity and developer tooling. Why would an AI agent choose Starknet over Arbitrum if the court is the only differentiator? The answer is: it wouldn’t, unless the court is deeply integrated into a specific AI agent framework. No such integration is announced.
Every line of code writes a history of power. In this case, the code is unwritten. The power history is blank. That is the most damning indictment.
Contrarian Angle: The Narrative Shell Game
Let me play the contrarian role I was born into. Perhaps the Internet Court is a brilliant piece of strategic marketing. Starknet, despite its technical superiority, has struggled to attract the same developer attention as Arbitrum and Optimism. AI agents are the hottest narrative in 2025. By associating Starknet with a futuristic “court for AI,” the team captures mindshare. The announcement costs nothing but a press release. Even if no code ever ships, the narrative has already been printed.
But this is exactly the kind of hype that erodes trust in the industry. We saw it in ICOs: projects promising “world computers” and “decentralized governments” that delivered nothing. We saw it in DeFi: protocols with $10 million TVL and no revenue calling themselves “the next JPMorgan.” The Internet Court fits this pattern. It is a concept wrapped in buzzwords, with no verifiable output. The contrarian question is: what if the lack of detail is intentional? What if they are not hiding anything because there is nothing to hide?
Furthermore, consider the governance of the court itself. Who controls the upgrade key? Who can change the arbitration rules? If it is a DAO, how are tokens distributed? The announcement implies a “decentralized” model, but without token distribution or governance parameters, it is likely a multisig in practice. Centralized control over a “court” is a single point of failure that defeats the purpose.
Governance isn’t a feature; it’s the foundation. A court without structural governance is a dictatorship with a friendlier name.
Truth emerges from transparency, not from silence. The silence in this announcement is deafening.
Takeaway: Wait for the Gray Paper
The Internet Court is not investable, buildable, or critiqueable today. It is a press release. As an analyst, my recommendation is to ignore it until three signals appear:
- Open-source code with a professional audit. No code, no trust.
- A real arbitration case with a verifiable on-chain outcome. AI agent disputes are rare; a single case would prove the mechanism works.
- Integration with a known AI agent framework like Autonolas, Fetch.ai, or SingularityNET. That would show actual adoption.
Until then, treat the Internet Court as a narrative placeholder. The real work of building justice for autonomous agents lies ahead, and it will require more than a press release. It will require rigorous engineering, economic modeling, and legal adaptation. The path to decentralization is paved with audited code, not with announcements.
We didn’t need another court. We needed verifiable justice. The Internet Court has not delivered it. But the conversation it sparks may accelerate the real innovation.
Now, let’s get back to work. Audit the intent, not just the syntax.