AI agents spend money online? New research shows that is not the case

TRM Labs reports: Most of the transaction volume of Coinbase's x402 protocol does not come from AI agents.

According to data released by blockchain intelligence company TRM Labs, most of the payment transaction volume in Coinbase's x402 protocol does not originate from artificial intelligence (AI) agents.

Data Insights and Estimation Bias

The TRM report released on Wednesday analyzed $52.7 million of 198.9 million settlements processed by known x402 intermediaries on the Base, Solana and Polygon networks since May 2025. After excluding self-payments and other abnormal traffic, the researchers estimated that only 0.6% to 7.5% of the value of the remaining business activity could come from AI agents.

report states: "Suppose a real agent explores among multiple services and products, while repeating addresses at the same price behaves more like a script that keeps clicking on a single service." "This was a deliberate modeling choice that could underestimate the actual scale of the field: many agents today may be dedicated, paying repeatedly for a single service, and such testing could misinterpret such behavior as script," the report emphasized.

x402 protocol mechanism and non-AI traffic interference

The x402 protocol was launched by Coinbase in 2025 to integrate payment functions into web requests. In this process, the buyer receives the price and signs a payment authorization; the intermediary verifies the authorization, submits the blockchain transaction, and pays the network fee.

However, TRM Labs points out that ordinary scripts can follow this sequence without an AI agent. Timed tasks, load testing, and self-transactions also produce the same on-chain records, so the total amount of agreements alone is not enough to accurately measure agency business activities.

In order to more accurately identify real transactions, TRM excluded self-payments, bulk flows from one or two payers, and sellers with fewer than 10 buyers, ultimately retaining $25.62 million in possible commercial transactions. Subsequently, the screening conditions were further tightened, requiring intermediaries to broadcast payment amounts varying and with an average value of less than $1. More stringent tests require the model to last for several months, or to have public agent registration information, or to make payments to multiple sellers. TRM warns that these standards may leave out real agents who purchase the same service repeatedly.

Capital flows and market dynamics

Data shows that activities at the end of 2025 include the minting of seemingly meme-tokens and payments for a single AI analytics service. In early 2026, transaction volume was concentrated in a single payment contract, and then around the middle of the year, AI service payments resumed through a proxy payment router. During the entire statistical period, USDC accounted for 99.6% of settlement value, or $524.7 million.

At the same time, major technology companies continue to invest in the field of agency payments. Binance's Agent OS launched in August includes an x402 payment layer;Coinbase's Base Network is specifically targeted at agent and payment startups through its $1 million accelerator program. In addition, Amazon announced in May that it has partnered with Coinbase and Stripe to launch AgentCore Payments, allowing agents to use stablecoins to pay for online services.

Regulatory Challenges and Compliance Recommendations

In addition to measuring usage, TRM also points to gaps in allocating agent payment responsibilities.

TRM Labs wrote: "The on-chain agent registry allows individuals to claim ownership of an agent address. However, such statements are voluntary and most participants currently do not take advantage of this feature."

TRM calls for the establishment of more accurate registration mechanisms, counterparty reputation information that agents can self-verify, and a monitoring system to accommodate massive amounts of micropayments. "The infrastructure is in place. What we need is accurate registration, counterparty reputation that agents can independently verify, and a monitoring system designed for high throughput rather than high value." They concluded,"Agency businesses require agent-level compliance."