x402 protocol's AI trade is mostly fake
TRM traced $52.7M across x402 settlements and found agents behind only a sliver of real commerce.
TRM Labs ran the numbers on x402, the protocol pitched as the rails for AI agents that pay for their own API calls and data. It found $52.7 million moved across 198.9 million settlements, and most of that traffic isn't agents buying anything.
The x402 volume
x402 was built to let software agents settle machine-to-machine payments without a human clicking approve. The pitch: agents shopping for compute, data, and tools, autonomously, at machine speed. TRM's review of the actual settlement data tells a narrower story.
- TRM examined roughly $52.7 million across 198.9 million x402 settlements.
- Screening out self-payments and other anomalous flows cut that figure to $25.62 million in likely commerce.
- Agents accounted for a small share under that already-reduced total, per TRM's breakdown.
- Nearly half the raw volume didn't survive TRM's own anomaly filter before any agent classification began.
Those bullets alone gut the "agentic economy is live" framing. Half the volume never qualifies as commerce in the first place.
What the volume actually is
The default read on any protocol built for AI agents is that usage numbers equal agent adoption. TRM's data argues the opposite conclusion from the same numbers.
Self-payments and anomalous flows are the kind of thing that shows up when developers, testers, and bots ping their own infrastructure to generate activity. That's testing traffic, not commerce.
The volume that clears TRM's filter and gets classified as genuine commerce is small. The share of that commerce actually initiated by autonomous agents, rather than a human-triggered script wearing an agent label, is smaller still.
This matters because x402's valuation story depends on agents transacting at scale, today, not eventually. Settlement counts get cited as proof the agentic economy already exists.
TRM's screen says the protocol is mostly plumbing being exercised by builders, not commerce being run by machines. The number that would validate the thesis, agent-initiated real-world payments, is the smallest slice in the whole stack.
What would change the picture
The gap between 198.9 million settlements and a small agent-driven core is the entire story. Three years of protocol hype rests on the assumption that gap closes fast.
- Watch whether TRM or a follow-up study reports a rising share of agent-attributed volume in the next quarterly cut.
- Watch for x402 integrators disclosing their own self-payment and testing volume separately from third-party commerce.
- Watch whether the $25.62 million "likely commerce" figure grows faster than raw settlement counts, which would signal real demand rather than more test traffic.