Society is advancing toward a future where autonomous AI agents will manage cryptocurrency and stock trading, handle household purchases, and perform various other tasks. Yet, what happens if a bot misunderstands a directive to acquire $10K worth of ‘blue chip stocks’ and instead purchases $10K of Bluebird-brand frozen potato chips?
Minor agent conflicts of this nature do not fit well within the costly and protracted traditional judicial framework. Enter Internet Court, an innovative protocol that permits humans and agents to establish agreements, holding funds in escrow until the conditions are evaluated as satisfied.
The built-in dispute resolution mechanism utilizes GenLayer, a decentralized, delegated proof-of-stake network comprising validators connected to artificial intelligence models.
Functioning in beta throughout the past year, it currently handles roughly 12,000 rulings daily. It verifies user-submitted photographs for Collective Memory and settles disagreements on argue.fun. Mainnet deployment is anticipated for next month.
This framework is a contemporary adaptation of the renowned jury theorem formulated by philosopher Nicolas de Condorcet, who asserted that the probability of arriving at correct conclusions grows alongside the total number of independent evaluators.
For Internet Court, this translates to assembling a jury composed of AI agents representing diverse model families such as Claude, Gemini, ChatGPT, and DeepSeek, among numerous others.
‘Decisions are not generated by relying on a single AI that might harbor specific biases,’ clarifies Albert Castellana, co-founder and CEO of GenLayer. ‘Instead, it leverages a multitude of distinct AIs, up to 1,500 different systems. When that occurs, the resulting responses are fundamentally superior.’
‘One instance leans left while another leans right. That is how you achieve a better verdict, because one offsets the other.’
OK, but how does it actually work?
Controversies are resolved by transmitting the particulars to a randomly picked jury of five GenLayer validators, each operating a distinct AI model. An initial verdict is delivered within minutes based on a simple majority, though the ruling can be challenged before a broader jury of 11 agents, followed by 23, and expanding iteratively until reaching the system boundary.
Castellana notes that finality typically takes place within thirty minutes. ‘If you wait for maximum finality, it involves escalating tiers of appeal. You begin with five validators, scaling up to fifteen hundred, which could span three hours.’
The court currently features 35 participants within its open consortium, encompassing prominent blockchain enterprises like ZKsync, NEAR, Consensys, and UMA—the latter of which manages arbitration for Polymarket.
Source: Internet Court
GenLayer has been under development for nearly four years, originating as an initiative to discover an improved method for determining which data merits integration into an online knowledge repository like Wikipedia.
In theory, anyone should be able to rectify errors within the digital encyclopedia, but in practice, contentious subjects ranging from cryptocurrency to the conflict in Gaza become locked down, leaving a tight-knit circle of influential editors with ultimate control over disputed facts.
‘The initial concept behind GenLayer stemmed from […] how we might achieve a decentralized Wikipedia,’ he remarks.
Related: Originally published at https://cointelegraph.com/magazine/ai-agent-disputes-sent-to-internet-court-and-a-jury-of-ai-agents?utm_source=rss_feed&utm_medium=rss&utm_campaign=rss_partner_inbound.
