AI Agents Made 6.4 Million Payments. Why Won’t Institutions Trust Them?
In May 2026, a public put up carrying Morse-coded directions brought about an AI-linked pockets to ship $175,000. Roughly 80% was recovered. The switch was legitimate on-chain, but the software program had accepted directions from a supply it ought to by no means have trusted.
The incident captures a significant impediment to autonomous finance: how do companies belief AI brokers with severe cash?
BeInCrypto Intelligence’s State of AI Agent Payments 2026 tracked 6.4 million x402 funds throughout Base and Solana in 35 days, price simply $119,947. More than 90% have been under one cent.
Five trade leaders spoke to BeInCrypto about what comes subsequent.
Millions of Payments, Little Money
x402 lets software program pay for on-line sources robotically. Yet its transaction rely says little about precise industrial demand. Some funds could contain testing or repeated service requests.
“Micropayments may stay nearly all of transactions whereas buying and selling and treasury operations account for a rising share of the worth moved,” mentioned Amir Forouzani, co-founder and CEO of Puffer Labs.
Saad Naja, founder and CEO of Hygentra, expects AI buying and selling to develop a lot sooner.
“I anticipate the subsequent 12 months to be a breakout 12 months for agent buying and selling, though micropayments will nonetheless dominate transaction rely. In 5 years I anticipate 80% of buying and selling will probably be agentic,” he mentioned.
That forecast is way forward of observable proof. Across 4 chosen Hyperliquid connections, AI-labelled functions accounted for simply 0.55% of buying and selling worth. Unlabelled AI exercise couldn’t be measured.
The Most Dangerous Payment Can Be Legitimate
The report recognized two incidents involving $281,200 transferred to attacker-controlled wallets earlier than recoveries. Neither required breaking a blockchain signature. The brokers acted on directions they need to have rejected.
“A securely saved personal key can nonetheless authorize a disastrous transaction if the agent has been manipulated by a malicious web site, bill or response from one other service… Spending limits alone are inadequate. You additionally want to regulate whose directions the agent accepts and implement its permissions independently of the mannequin,” Forouzani warned.
One agent’s mistake may additionally unfold throughout many others utilizing the identical data.
“When hundreds of brokers purchase the identical cheap feed and act on it at machine pace, one dangerous enter will get repeated throughout all of them without delay, and a spending cap set per agent does little or no to include that,” mentioned Markus Levin, co-founder of XYO.
A Transaction Can Be Reversed After an Agent Acts
Benedikt Bünz, co-founder and chief scientist at Espresso Systems, says brokers face one other threat once they spend funds earlier than an underlying transaction is closing.
“Ethereum transactions are solely closing after 16 minutes, i.e., they might be reverted earlier than that. Agents both want to take a seat idle for that point or make their subsequent transfer on a cost that would nonetheless be reversed. This publicity can have harmful trickle-down results at scale,” he mentioned.
Automated trades and funds typically depend upon earlier transactions. A reversal can have an effect on later steps. Ethereum’s finality course of reveals why affirmation and finality matter.
Who Takes Responsibility When AI Gets It Wrong?
In the report’s readiness framework, micropayments scored 60.7 out of 100, whereas institutional automation scored 12.0. Limited disclosure of real-world outcomes contributed to the low institutional rating.
Across 19 merchandise reviewed, the 14 that dealt with cash stayed inside human-set limits.
Chandler Fang, co-founder of t54 Labs, sees an unresolved query:
“Now brokers have the technical functionality to maneuver cash round and typically make their very own choices. How will we ensure that when brokers make the improper transactions, there are clear procedures and protocols to deal with the draw back threat?”
Levin expects the subsequent main alternative to come back from physical-world funds.
“The subsequent class of agentic funds I’d anticipate to develop is funds launched as soon as work is completed within the bodily world, comparable to a supply being accomplished or a job completed on web site, and people carry a lot bigger quantities than an API name,” he mentioned.
Such funds require proof {that a} job was accomplished, permission guidelines enforced outdoors the AI mannequin, and clear legal responsibility when a transaction goes improper. Institutions nonetheless want proof that these safeguards work at scale.
The put up AI Agents Made 6.4 Million Payments. Why Won’t Institutions Trust Them? appeared first on BeInCrypto.
