When AI learns to spend money, how should we respond?

When AI can no longer just think and express itself, but is able to "spend money" autonomously, how will the world's economic order be rewritten?

Author: Liu Honglin, Attorney at Law, Mankiw Blockchain Legal Services

Over the past decade, AI has been defined as a form of "intelligent productivity"—helping humans make decisions, optimize processes, and save costs. But now, a more fundamental question is looming: when AI can no longer just think and express itself, but autonomously "spend money," how will the world's economic order be rewritten?

The emergence of AI payments signifies the first time machines have participated in value exchange. It transforms from a cognitive system into an economic entity. It requires identity, accounts, credit, contracts, settlement, and even responsibility. When discussing the next step for AI, the focus shifts from algorithms, computing power, or large-scale models to more fundamental questions—when AI truly participates in economic activities, how will it "pay"? Who will endow it with accounts, credit, settlement, and responsibility?

The redesign of the payment system will determine the boundaries of the intelligent economy. We have found that three distinct paths have emerged regarding "how AI pays."

One approach is an institutional path centered on identity and regulation , attempting to allow AI to operate legally within the existing financial system; another is a crypto-native path based on blockchain , enabling AI to become an autonomous economic node; and yet another is a vertically integrated path centered on the enterprise ecosystem , using productization and encapsulation to bring AI payments to fruition.

These three paths are like the Trisolaran civilization, independent yet interacting with each other. They are not mutually exclusive, but rather, under different trust logics, they explore the same question: in a world where machines can also fulfill contracts, how should trust be rebuilt?

Google: Let AI "spend money on your behalf" within the system.

Google's approach most resembles imperial thinking. Its AP2 (Authorized Payment Protocol) is an "AI-authorized payment protocol" that attempts to embed machine economic activity into the existing financial system . In other words, it doesn't aim to give AI accounts, but rather to allow AI to execute payments on behalf of humans within authorized scopes.

Its core mechanism consists of three parts: identity verification, authorization credentials, and payment execution.

Users first define their AI's permissions within the system—for example, this AI can pay subscription fees, run ads, and settle computing power bills on my behalf, but cannot transfer money or make investments. Whenever the AI initiates a payment, the system generates a short-term encrypted authorization (similar to a one-time digital signature), which is verified by Google's network before the transaction is executed. Once the payment is complete, the authorization automatically expires.

Under this mechanism, AI is merely an executor, not the account holder. The money it spends still comes from human wallets, and settlement still goes through payment channels such as banks, card organizations, or Google Pay. AI is just an "agent," and its authority and boundaries are jointly defined by users and regulators.

The problem Google wants to solve is: how can AI safely help people spend money without overstepping its authority or breaking the law? It's not concerned with freedom, but rather with controllability and compliance . This is "imperial trust": trust comes from identity and regulation, not algorithms.

Coinbase: Let AI "spend its own money"

Coinbase takes the opposite approach. Instead of letting AI represent anyone, it allows AI to become an economic agent on its own .

In this crypto-native system, each AI can generate its own crypto wallet (equivalent to a bank account), and its behavior rules are defined by smart contracts.

For example, you can pre-deposit 1 ETH into this AI wallet and write rules such as: "Each task expenditure must not exceed 0.05 ETH; payments must be recorded in an on-chain contract; if the balance is below 0.1 ETH, automatically request additional funds." From that moment on, this AI becomes an independent "economy." It can autonomously trade with other intelligent agents, pay API fees, purchase computing power, and distribute rewards— the entire process requires no human authorization.

Coinbase's x402 protocol is designed for this autonomous economy, allowing direct communication and settlement between different AIs. Transactions are completed via the blockchain, with funds transferred directly from the AI's wallet, without the need for intermediaries such as banks or payment companies.

The core logic of this system is that trust no longer comes from identity, but from code. There is no superior authorization, nor human review—contract rules are the law. It addresses another problem: how can AI independently participate in economic activities without human endorsement? This is also why it is a major concern for regulators: in such a world, who owns the assets? Who bears the risks? Who can regulate?

Stripe: Implementation is paramount.

If Google wants AI to spend money within the cage of regulations, and Coinbase wants AI to break free and spend money on its own, then Stripe represents a third way of thinking—enabling AI to spend money first. It doesn't talk about grand institutional revolutions, nor does it pursue freedom in the crypto world; instead, it returns to the starting point of commerce: settlement .

Stripe's starting point is actually quite practical. In today's internet world, almost all economic activities are based on "automatic payments"—ad placement, subscription fees, API calls, cloud computing power settlements—none of which can be completed through manual instructions. For AI to truly enter the commercial cycle, it must have the ability to "autonomously settle payments." However, traditional payment systems do not support this, and encrypted systems are not recognized by regulators. Therefore, Stripe chose a third path—building a usable bridge between the existing system and the future system.

It established an enterprise payment network called Tempo , which is neither reliant on the authorization system of financial institutions like Google, nor completely decentralized like Coinbase. Instead, it is custodied and audited by Stripe itself . Tempo is closed, but it is compatible with blockchain-style smart contracts and can also connect to fiat currency accounts. Developers only need to call Stripe's SDK, and the AI can complete actions such as payment, settlement, and tax filing in the background—for example, it automatically pays advertising budgets, purchases computing power, and settles data fees, all of which are executed by Stripe's system in the background.

In this model, AI neither owns the wallet nor signs authorizations; it relinquishes payment power to the platform. Trust no longer comes from regulation or algorithms, but from the company itself—from Stripe's credit, compliance, and risk management. This is a classic example of "business trust": not because the world is perfect, but because someone is willing to take responsibility.

The differences become even clearer when viewed in specific scenarios. Consider an AI managing an advertising account: In Google's system, it requires authorization before making payments, and its spending limits and uses are within regulatory rules; in Coinbase's system, it owns its own wallet and can directly settle advertising fees on-chain, with transactions public but irreversible; while in Stripe's system, it doesn't care about wallets or signatures, it simply issues commands—the Tempo network automatically handles all clearing, taxation, and compliance reporting. AI payments become like calling a function—clean, fast, and transparent.

Stripe's challenge isn't whether AI can spend money, but rather how AI can safely spend money within the real-world financial system. It chooses to use corporate credit escrow machines to encapsulate complex financial relationships in a productized way. However, the cost is clear. Tempo is Stripe's private network, with all settlement paths controlled by the company. If the platform malfunctions, the entire system could shut down.

If Google represents the extension of the system, and Coinbase represents the challenge to the system, then Stripe represents the integration of the system—replacing institutional conflict with business efficiency. Its revolution is not about overthrowing the old world, but about enabling new technologies to take root within the old world first. The competition among these three is not only a competition of technology, but also a competition of systems. Ultimately, who will be responsible for the future AI world? Will it be the regulators, the code, or the platforms?

This won't be an either-or choice. There's never just one winner in the payment system. Whether it's VISA, SWIFT, PayPal, or the digital yuan, they've all coexisted for decades. The emergence of a new payment system doesn't eliminate the old one; rather, it coexists and overlaps with it until it's accepted by both the market and regulators.

Legal challenges

In this transformation, the real challenge lies not in technology, but in law.

Can AI become an independent economic entity? Can it legally hold assets and sign contracts? If its payment behavior errs, who should bear the responsibility? These are the fundamental questions facing all current AI payment solutions. In the on-chain world, smart contracts are irrevocable, which strengthens trust but also makes errors more fatal. Once AI misjudges and funds are mistakenly transferred, there is no "reversal" button. In traditional legal systems, the traceability of transactions is the bottom line—every payment must have a clearly defined responsible party. If AI's autonomous payments cannot be incorporated into the legal liability framework, its "freedom" cannot be recognized by the system.

This is why I believe AI will not possess true economic personhood in the short term. It can execute payments, but it cannot bear the consequences. Every step of AI payment still requires human signatures, platform escrow, and institutional endorsement. This is similar to the early stages of self-driving cars; technically, we can relinquish control, but legally, we dare not. The future of AI payments is similar—we must first design a complete "intelligent agent liability system," including authorization rules, loss compensation, risk insurance, and regulatory interfaces; otherwise, this system will quickly collapse.

In the short term, AI payments will initially appear in existing payment systems as "smart authorization," such as the auto-settlement function in Google Pay, Apple Pay, or WeChat Pay. In the medium term, enterprise-level scenarios (SaaS calls, advertising settlements, API billing) will be the first to form an automatic payment ecosystem, with the Stripe model being the most likely to be commercialized. In the long term, a decentralized system like Coinbase, while facing the greatest compliance pressure, is most likely to foster genuine institutional innovation because it raises the most fundamental questions—who owns the assets, who defines trust, and who bears the responsibility.

Technology will ultimately force the law to provide new answers. Perhaps future contract law will add a "smart agent liability clause," or perhaps future anti-money laundering regulations will include a chapter on "AI customer identification." The evolution of AI payments will ultimately drive institutional evolution.

Disclaimer: As a blockchain information platform, the articles published on this site represent only the personal views of the authors and guests and do not reflect the position of Web3Caff. The information contained in the articles is for reference only and does not constitute any investment advice or offer. Please comply with the relevant laws and regulations of your country or region.