Who Will Regulate AI Agent Disputes? Blockchain 'Internet Court' Has Opened

In the near future, everyone will use AI agents. These agents connect to the cloud, reside on smartphones, and can help you reply to emails, book flights, and even automatically manage tax-loss harvesting for investment portfolios.

Written by: Forbes
Compiled by: AididiaoJP, Foresight News

Robinhood customers are now using AI agents to analyze stock market fluctuations and trade autonomously based on personal instructions. SAP's Joule helps enterprise clients analyze inventory, find the best suppliers, and complete purchases. Shopping agents like Amazon's 'Buy for Me' scan the web for optimal deals at machine speed, negotiate terms with sellers, confirm delivery times, and make direct payments.

From Anthropic and OpenAI to well-known cryptocurrency companies like Coinbase and Circle, many are pushing for this agent-driven future to arrive sooner.

But what happens when the sofa ordered by the agent is the wrong color, arrives two weeks late, or is damaged while the seller insists it was a post-delivery issue? This seemingly trivial yet potentially costly problem lurks within the grand blueprint of agent-based commerce. While software can represent humans and businesses in shopping, bargaining, hiring, and payments, AI can sometimes 'hallucinate,' and commerce is never just about money; accidents will always happen.

"Agent-based commerce is at a critical turning point, and we are not yet prepared for the potential consequences," says David Riudor, CEO and co-founder of the GenLayer Foundation. Based in the Cayman Islands, the foundation operates a new blockchain called GenLayer and its core application—the Internet Court. This court is designed specifically to adjudicate disputes between agents and operates without human intervention, having garnered support from 26 cryptocurrency and AI companies, including the crypto exchange OKX, wallet provider MetaMask, and Binance's BNB Chain.

The Internet Court does not aim to completely replace human judges with robots but rather to help agents pre-sign contracts with clear terms. When both parties cannot agree on the outcome, an AI jury evaluates the evidence and delivers a verdict within minutes.

Riudor points out that this technology is particularly suited for small transactions—hiring a lawyer is not cost-effective, but ignoring the issue can lead to losses. "We are not trying to compete with the traditional legal system," says Albert Castellana, co-founder and CEO of GenLayer Labs. "We are simply providing an alternative. For a $10,000 claim, hiring a lawyer is not economical; this system can resolve it quickly, potentially costing just a few cents."

The potential market size is enormous. According to Adobe Analytics, since October 2024, retail website traffic driven by AI recommendations has increased by over 14 times. McKinsey predicts that by 2030, AI agents could facilitate $3 trillion to $5 trillion in global consumer commerce. However, most emerging infrastructure still focuses on the 'smooth path'—agents finding what their owners want, completing payments, and moving on after receiving goods or services.

Currently, the Internet Court is applied in limited scenarios. The social platform Collective Memory rewards users for uploading real-time photos, videos, and reports. When it needs to determine whether a disputed image is fake, it calls upon the GenLayer system. For example, a video of a bombed school in Gaza or footage from the streets of Tehran will be assessed by the Internet Court, which will consider the upload time, location, relevant submission records, and user activity history as evidence to determine authenticity.

The ultimate goal is for the Internet Court to automatically intervene when AI agents dispute with each other.

Imagine a small online clothing company where the owner has delegated daily operations to multiple agents: one manages inventory, another buys ads, and another finds creative talent. The owner wants a new logo, and the agents find a designer represented by another agent. They agree on design, price, and delivery date, and when the logo arrives, it looks good but is found to potentially plagiarize another's portfolio through reverse image search.

The design of the Internet Court is precisely for such situations: allowing agents to pre-agree on terms, escrow payments, and submit disputes to the jury before funds are released.

The core of the jury is blockchain technology. It consists of five randomly selected blockchain validators, each running different models (e.g., Claude, GPT, Gemini). One serves as the leader to propose an initial decision, while the others vote secretly before publicly stating whether they agree. If consensus is reached, a 30-minute dispute window opens, during which agents or humans can challenge by posting a bond. If challenged, the jury expands to 11 members, and so on, until consensus is formed and no further objections are raised.

This mechanism draws on the jury theorem proposed by Enlightenment philosopher and mathematician Nicolas de Condorcet in 1785: the more independent evaluators there are, the higher the probability of reaching the correct conclusion. GenLayer believes that a combination of multiple models is harder to manipulate than a single model or a single human arbitrator.

Although disputes between agents may sound futuristic, the Internet Court is already online and in testing. According to Castellana, the network processes about 350,000 transactions daily, resulting in 20,000 to 25,000 decisions. A formal public launch is planned for later this year, along with a token issuance to attract more validators—anyone can participate in this role.

Riudor states that this system is not limited to agent-based commerce but can also be used in prediction markets. For example, Polymarket currently relies on the UMA protocol to let token holders vote on dispute outcomes, while AI-assisted adjudication would be faster. "We are in talks with some large prediction markets," Castellana says, "they are waiting for us to fully launch but are already evaluating."

Andrew Hall, a professor at Stanford Graduate School of Business and a research advisor for a16z's crypto team, noted earlier this year that using large language models as adjudicating judges could help scale prediction markets because the models cannot be bribed and their performance is rapidly improving. However, he also cautioned that models can still hallucinate and may be manipulated through clever prompts or contaminated training data.

Lindsay Lin, former legal advisor and now COO of the New York crypto venture capital firm Dragonfly, also sees similar issues: "Many large language models share training data and common flaws, leading to correlations, while human judgment tends to be more independent. However, people are likely to prefer AI for low-value disputes because it is cheaper and faster than human jurors. The scale of agent-based commerce is vast, and the volume of disputes will also be large. Having standardized agreements for agents makes sense so that they are clear on the terms of cooperation and remedies when transactions are not completed correctly."

Others have reached similar conclusions. Just two weeks ago, the largest arbitration institution in the world—the American Arbitration Association's International Dispute Resolution Center—announced the launch of a 'Legal Context Protocol' standard for agents. This protocol is co-led with Denver blockchain company Integra Ledger, with founding contributors including Google, IBM, and several leading crypto companies like Circle and Ava Labs.

Of course, whether these standards can gain widespread adoption ultimately depends on broad acceptance and whether AI models can effectively reduce hallucinations and biases.

Meanwhile, the infrastructure for agents to find, hire, and pay each other is rapidly taking shape. In recent weeks, GenLayer partner OKX and the AI-focused NEAR blockchain team have launched markets where agents can hire other agents to complete paid tasks, such as obtaining datasets or assisting with code reviews.

Real courts have also begun handling cases of AI agent violations. In November 2025, Amazon sued Perplexity, accusing its AI-driven Comet browser of logging into customer accounts and masquerading as a regular Google Chrome browser to shop on the Amazon platform without authorization. In March, a federal judge in California issued a preliminary injunction prohibiting Comet from shopping on Amazon, but the appeals court later stayed that order and is reviewing Perplexity's appeal.

Regardless of how the court ultimately rules, this case highlights a larger challenge facing agent-based commerce: when millions of AI agents act on behalf of users across platforms, how can effective regulation be achieved without a unified enforcement mechanism?