BeInCrypto publishes AI agent payments report at TOKEN2049

Drawing on input from the BeInCrypto Expert Council, which includes experts from AWS and Cloudflare, the new report investigates observable activity, the amount of money in motion, and where the hype goes beyond the data.

BeInCrypto Research compiled the report.

Autonomous AI agents that can make payments are among the most discussed topics in crypto at present. Stablecoins, emerging payment protocols, and high-profile launches consistently generate headlines. However, the question remains: how much of this is real activity?

That question drives "The State of AI Agent Payments 2026", a new report from BeInCrypto Research. The report is being launched at TOKEN2049 Singapore, which runs from 7-8 October at Marina Bay Sands.

The report examines real-world usage rather than predictions. It traces funds on-chain, assesses what products enable agents to accomplish, and poses a straightforward question for businesses: what conditions must be met for this to reach mainstream adoption?

What the research uncovered

• Most x402 payments are very small. Over a 35-day sample of x402 payments on Base and Solana, 6.4 million transactions totaling only $119,947 were recorded. 90.8% of these transactions had a value below one cent. This represents the type of payment that payment cards were never designed for, which clarifies why stablecoins are frequently discussed.
• A high volume of transactions does not indicate high demand. Solana contributed 94.5% of the sample's transactions, with most occurring over only eight days. On 26 August, 145 addresses sent 1,283,926 payments to just three recipient addresses. The report does not characterize this as wash activity or assert intent, but it demonstrates why transaction numbers alone are not meaningful for assessing actual business demand.
• Payments are for small, practical items. The most frequently requested paid x402 resource was a token safety check costing two cents, which received 63,319 calls during the catalogue's sampled period. The majority of the top services provided information or executed minor software tasks. These figures represent recorded calls, not confirmed purchases.
• Humans remain in control. Among 19 AI products and workflows reviewed, 14 operate within pre-defined limits set by a person or company. None of the 19 establishes its own objectives. Five out of the seven trading platforms examined explicitly prevent agents from withdrawing funds.
• AI-related trading represents a minor portion. Across four chosen Hyperliquid connections, applications categorized as AI processed only 0.55% of $653.08 million in trading value. Nearly all, 99.4%, was routed through CCXT, a standard connection tool without AI claims. AI may also trade via CCXT without being labeled, so this is not a comprehensive market figure.
• Failures occur quickly. Two incidents were documented, including one where a public message prompted a $175,000 transfer from an AI-linked wallet. Combined, $281,200 was sent to attacker-controlled wallets before any funds could be recovered.
• Investment is flowing into infrastructure. 36 funding rounds and acquisitions with a disclosed value of $4.25 billion were reviewed. Approximately 92% went to payment and cash-management firms, rather than to trading agents.