x402 Settled 200 Million Agent Payments, but Most Were Worth Under a Cent

Since its launch, x402 — a Coinbase protocol designed for agent payments — has processed roughly 200 million transactions. AI optimists frequently cite that figure as evidence that machines can manage money without people. Joscha Bach, a German scientist and AI researcher, has said much the same.

Yet how far does that claim hold up? According to BeInCrypto’s State of AI Agent Payments 2026 report, the bulk of agentic payments carry a value below one cent.

Over a 35-day x402 sample, agents completed 6.4 million transactions while transferring a combined $120,000.

Does that leave the optimism around AI drifting too far from what is actually happening?

What if there is no hard problem?

— Machine Learning Street Talk (@MLStreetTalk), November 2, 2026

Does Money Need People at All?

For Bach, money is a pattern of cause and effect that owes nothing to paper or ink. Software, in his view, is able to carry that pattern by itself.

“It can play a stock market entirely without human intervention. And so arguably, money does not actually need humans to exist once you bring it into existence as long as you have machines that can deal with money.”

The BeInCrypto State of AI Agent Payments 2026 report shows that 90.8% of those payments were valued at under one cent.

Information was the main thing purchased, with token safety checks and web searches ranking among the most requested services. Rather than a boom, the findings describe a reality check for the agent economy.

Why Human Budgets Still Govern Agent Payments

Nineteen AI products and financial workflows, ranging from exchange tools to bank systems, came under review. Fourteen of them could operate only inside limits a person had approved beforehand.

Not one allowed an agent to set its own objectives. On top of that, five of the seven trading venues barred agents outright from withdrawing funds.

Trading figures point the same way. On Hyperliquid, apps tagged as AI accounted for just 0.55% of the $653 million traded between Jan. 1 and Aug. 25.

So the machine-run money Bach describes does exist in principle. What stays with people, in practice, is the budget — which explains why spending limit risks sit at the centre of the business debate.

When a Zombie Holds the Keys, What Then?

A chess program, in Bach’s terms, is a zombie: skilled at its task yet devoid of any inner experience. An agent that pays for data would probably meet the same standard.

Bach suspects consciousness is a learning algorithm through which a brain monitors its own perception. Neuroscientist Anil Seth, whom Bach disputes, holds instead that consciousness may rest on living biology.

Whichever view is right, today’s agents have no such self-check, and losses demonstrate it. In May 2026, an attacker posted a Morse code message that Grok, xAI’s chatbot, translated.

Bankrbot, which settles payments from wallets it is linked to, read the reply as an instruction and dispatched $175,000 in tokens. Roughly 80% of that was later returned following the Morse code exploit.

In a separate incident, an attacker got into the dashboard of AIXBT, an AI agent known for publishing market commentary. The breach diverted 55.5 Ethereum (ETH), worth $106,200 according to the report.

“No one queries a blockchain directly… Unlike a human, an agent will never stop and think that number looks off.”

— Rodrigo Coelho, CEO of Edge & Node, in BeInCrypto’s report

Coelho further cautions that when many agents draw on the same source, they can all make the same mistake at once. That echoes Bach’s observation that separate AI models tend to acquire remarkably similar internal features.

As things stand, the report identified no product in which an agent determines its own goals. Should that boundary ever be crossed, the consciousness debate would turn from philosophy into a question of liability.