x402 Shows the Gap Between AI Payments and Human Budgets
Research into Coinbase’s x402 shows that AI agents mostly make microtransactions, while humans still set budgets and trading limits.
Key Takeaways
- Coinbase’s x402 has processed about 200 million transactions since launch, but most payments remain extremely small.
- Over 35 days on x402, agents made 6.4 million transactions worth just $120,000; 90.8% were worth less than a cent.
- The research shows humans are still in control: agents are not allowed to choose goals on their own and usually operate within preapproved limits.
x402, Coinbase’s protocol for payments made by AI agents, has processed about 200 million transactions since launch. Still, new research shows that this machine economy is much smaller for now than the hype suggests. In practice, most payments remain extremely small, while humans still set the boundaries and budgets.
Small Payments Dominate
BeInCrypto’s State of AI Agent Payments 2026 report shows that 90.8% of agent payments were worth less than a cent. In a 35-day sample on x402, agents made 6.4 million transactions, but together they moved only $120,000 (€107,300). That suggests most of the activity is about small, functional payments rather than large autonomous money flows.
The agents mostly bought information, while token safety checks and web searches were also commonly purchased. The report looked at 19 AI products and financial workflows, from exchange tools to banking systems. In 14 of them, an agent could only act within preapproved limits.
Humans Still Hold the Controls
According to the report, none of the systems studied were allowed to choose goals on their own. Five of the seven trading platforms also explicitly blocked agents from withdrawing money. On Hyperliquid, AI-labeled activity was also only visible in a limited way: it accounted for 0.55% of $653 million (€584 million) in trading between January 1 and August 25.
That fits the broader security concerns around AI in financial systems. Recent incidents have shown that AI agents can be vulnerable when they get access to payment or network environments, and major banks have already taken steps to lock down third-party access more tightly. For crypto, that matters because wallets and trading platforms are increasingly experimenting with automated agents, but control still sits with humans.
The debate around protection and payment reversals is also still active at x402 itself, because the protocol does not have a standard chargeback mechanism. That makes mistakes or abuse by agents especially hard to fix.
Why This Matters for Crypto
The discussion is not just about technology, but also about risk and responsibility. If an agent makes a mistake or is given the wrong instructions, it is still unclear who exactly pays the price. That makes the combination of AI, wallets, and crypto payments especially sensitive, especially now that the industry is innovating faster than many security and oversight frameworks can keep up.