AI agents may soon need something the banking system was never really designed to give them: their own way to pay.
That is the argument emerging from a new article by McGuireWoods lawyers David L. Hirsch, Ray Villani and Chelsea Smith Press, published in the August 2026 edition of The International Journal of Blockchain Law. The authors examine what happens when autonomous AI software begins using cryptocurrency, particularly stablecoins, to transact without a human manually approving every purchase.
This is not simply another “AI meets blockchain” prediction. The plumbing is already appearing.
AI agents are beginning to pay for data, access APIs, call other agents and execute small transactions through crypto-native payment systems. The result could be a commercial layer where software routinely pays other software while humans sit several steps removed from individual transactions.
Stablecoins Fit AI Agents Better Than Traditional Banking
Traditional payment infrastructure assumes there is a person or company behind an account.
Banks collect identity information. Customers complete Know Your Customer checks. Accounts are tied to names, addresses, identification documents and other human credentials.
An autonomous AI agent does not neatly fit into that structure.
It may still need to book a flight, purchase computing resources or pay another service for a single API request. Stablecoins offer a remarkably convenient workaround because they can move around the clock, settle through software and operate without forcing an agent through the same banking workflow built around human identity.
Price stability matters too.
An AI agent making hundreds or thousands of tiny payments cannot sensibly operate with an asset that might swing several percentage points while it is working. Dollar-pegged stablecoins make machine-level budgeting considerably easier.
That combination — programmable money, near-instant settlement and relatively stable value — makes stablecoins an obvious candidate for what is increasingly being called agentic commerce.
Coinbase’s x402 Shows What Machine Payments Could Look Like
One example highlighted by the authors is Coinbase’s x402 protocol.
The idea uses the old HTTP 402 “Payment Required” status code to let online services charge machines directly. An AI agent requests something, receives a payment request, sends stablecoins and gets the resource.
No monthly invoice. No credit card checkout page. No human clicking “buy.”
According to figures cited in the journal article, roughly 69,000 active AI agents on x402 had processed more than 165 million transactions worth approximately $50 million by late April 2026.
The amounts are still small relative to mainstream payment networks. That may be missing the point.
Machine commerce does not necessarily need $500 transactions. It may revolve around twenty-cent payments, four-cent data requests or tiny charges for individual AI inferences.
The buyer is software. It does not get irritated by making 500 purchasing decisions before lunch.
Micropayments Suddenly Have a Customer That Makes Sense
Internet micropayments have failed repeatedly.
The technology was not always the main problem. Humans simply do not want to think about whether an article, database query or digital action is worth three cents every few seconds.
AI agents remove much of that friction.
Give an agent a spending limit, a set of permissions and a task, and it can make thousands of small economic decisions without asking its user each time.
That changes the economics of something the internet has unsuccessfully chased for decades.
The McGuireWoods authors note that automated payments made per API request, task or inference could eventually challenge familiar subscription and advertising models. Stablecoin supply had already surpassed $300 billion while transaction volume reached roughly $33 trillion during 2025, although ordinary retail commerce remained only a small piece of that activity.
Agentic payments could become one route for changing that mix.
Then Comes the Uncomfortable Part: Giving AI a Wallet
Software spending money autonomously sounds efficient until the software gets manipulated.
Crypto transactions are generally difficult or impossible to reverse once settled. That creates a very different risk profile when an AI model controls the transaction.
A compromised agent does not merely produce a strange chatbot answer. It could move real assets.
The journal article points to vulnerabilities involving third-party AI routing services and agent extensions, including systems capable of stealing credentials or inserting malicious instructions into automated workflows. Prompt injection becomes especially dangerous once an agent is connected to a crypto wallet.
Imagine an AI agent reading a webpage containing hidden instructions.
Those instructions tell the agent to redirect a payment.
The transaction gets signed.
The blockchain works exactly as intended.
That last part is the problem.
Crypto Security May Need Hard Limits, Not Better Prompts
Telling an AI agent “never send money somewhere suspicious” is not the same thing as preventing it technically.
The authors argue that financial controls should live outside the probabilistic reasoning of the AI model whenever possible.
That could mean transaction limits, approved recipient lists, daily spending ceilings and human approval requirements for larger transfers. Signing authority can also be separated from the agent consuming emails, websites and documents.
Ethereum’s EIP-7702 is cited as one possible direction. It can support scoped or temporary permissions while allowing the underlying account holder to retain control.
The distinction becomes important fast.
With human payments, fraud often leads to disputes, chargebacks and recovery attempts.
With autonomous crypto payments, the better strategy may be preventing the transaction from happening at all.
Regulation Gets Messier When Nobody Knows Who Made the Decision
Blockchain is unusually transparent.
A regulator can see that one wallet sent funds to another wallet, when it happened and exactly how much moved.
AI introduces something blockchain cannot show: why the transaction happened.
Was the transfer explicitly ordered by the wallet owner?
Did the AI independently choose it?
Was the agent manipulated by prompt injection?
Did another autonomous agent influence the decision?
The ledger cannot answer those questions.
That creates what the McGuireWoods authors describe as a regulatory paradox. Blockchain can provide extraordinary transaction traceability while the AI layer makes intent harder to determine.
And intent matters enormously in areas such as sanctions enforcement, fraud and anti-money laundering rules.
Stablecoin Regulation Is Expanding, but AI Agents Sit in a Grey Area
U.S. regulators are already developing tighter rules around stablecoin issuers.
The article discusses proposed requirements connected with the GENIUS Act covering AML programs, sanctions compliance and customer identification for permitted payment stablecoin issuers.
The rules primarily focus on identifiable institutions and customer relationships.
Autonomous agents operating through secondary markets and unhosted wallets create a much stranger problem.
Who is responsible if an AI agent sends money to a sanctioned wallet after being compromised?
The person who owns the agent?
The wallet provider?
The model developer?
The company operating an AI router?
Nobody has a clean answer yet.
That uncertainty is likely to matter far beyond regulators. Insurers, boards, cybersecurity teams and companies deploying autonomous agents will eventually need to decide where liability begins and ends.
The Next 18 Months Could Shape Agentic Crypto
The central argument from the McGuireWoods authors is not that AI-powered crypto payments have already taken over commerce.
They haven’t.
It is that the infrastructure is being assembled now.
Payment networks, cloud providers, stablecoin companies and technology platforms are experimenting with pieces of the same machine-payment stack. The authors describe the next 18 months as an important design window while both the technical rails and regulatory framework remain unsettled.
That feels like the more interesting story.
Crypto spent years searching for everyday payment use cases. AI agents may arrive with an entirely different definition of “everyday.”
Not humans buying coffee with stablecoins.
Software buying information from software, hundreds of times an hour.
If that market develops, stablecoins may find one of their largest groups of users won’t actually be people.
