GenLayer, an open delegated proof-of-stake network, has launched the beta version of Internet Court, a protocol designed to resolve disputes between AI agents and humans using jury systems composed of artificial intelligence models. The system employs up to 1,500 different AI validators from families including Claude, Gemini, ChatGPT, and DeepSeek to reach majority-vote decisions on contract terms and subjective judgments. Currently processing approximately 12,000 decisions daily for platforms like Collective Memory and argue.fun, the protocol aims to provide faster, cheaper adjudication than traditional legal systems or existing crypto dispute mechanisms.
The core technology enables "intelligent contracts" that can interpret natural language, images, and code to make automated judgments, such as assessing content quality before releasing escrowed funds. Dispute resolution begins with a randomly selected jury of five validators, with options to appeal to larger juries of 11, 23, or more agents, potentially escalating to the full validator set. CEO Albert Castellana states that finality typically occurs within 30 minutes, though maximum escalation could take three hours. This approach contrasts with slower processes seen in prediction markets, such as the nine-day dispute over a $240 million Polymarket bet regarding Ukrainian President Volodymyr Zelensky’s attire.
Internet Court operates through a consortium of 35 members, including ZKsync, NEAR, Consensys, and UMA. While GenLayer originated from efforts to create a decentralized knowledge base similar to Wikipedia, its current focus is on autonomous agent interactions. Mainnet launch is scheduled for next month. The system allows for human appeals via Kleros, UMA, or traditional courts if users disagree with AI verdicts, acknowledging that AI judgment capabilities remain limited for complex tasks like deepfake detection or boundary disputes.
The introduction of AI-juried dispute resolution marks a significant shift in how smart contracts handle subjective conditions, moving beyond binary code execution to probabilistic reasoning. By leveraging diverse model families, GenLayer attempts to mitigate individual AI biases through collective decision-making, addressing a critical bottleneck in autonomous agent economies where traditional legal recourse is too slow and costly for micro-transactions. However, the reliance on statistical convergence among models poses a structural risk; if major AI providers align their training data or safety protocols, the diversity of the "jury" may diminish, leading to systemic blind spots rather than balanced adjudication.
Regulatory acceptance remains the primary hurdle for this infrastructure. While technically capable of resolving low-stakes disputes quickly, the legal enforceability of AI-generated verdicts is untested in most jurisdictions. The system’s design, which permits escalation to human courts, suggests it functions best as a preliminary filter rather than a final authority. Market participants should monitor whether institutional adoption grows alongside improvements in model transparency, as the credibility of these automated judges hinges on verifiable independence and resistance to adversarial manipulation.
