This article was first published on AI News on October 21, 2026. Click here to read the original article.
Artificial intelligence is rapidly moving beyond generating text and answering questions. AI agents are beginning to execute tasks, interact with external services, and make decisions with increasing independence. But as these systems become more capable, a fundamental question is emerging: what happens when an AI agent needs to spend money?
Allowing software to initiate payments, purchase data, or manage financial assets introduces challenges that go far beyond technical capability. Who authorizes a transaction? How are spending limits enforced? And who is accountable when an autonomous system makes a costly mistake?
These questions are driving growing interest in agentic finance, where AI systems operate within defined financial permissions and increasingly interact with blockchain-based payment infrastructure. Technologies such as the XRP Ledger, stablecoins, and emerging payment protocols like x402 are being explored as potential building blocks for this new economy.
To better understand what it will take to make autonomous financial transactions practical and trustworthy, we spoke with Odelia Torteman, Head of Digital Assets at XRPL Commons. She discusses the challenges facing AI-powered payments, the safeguards needed to balance autonomy with human oversight, and how the upcoming XRPL Hackathon in New York aims to bring these ideas closer to real-world adoption.
Q1. What is currently the single biggest obstacle preventing AI agents from making financial transactions independently, and how could blockchain infrastructure help address it?
Agentic finance is advancing tremendously and opening new opportunities across financial services and other industries, yet agents transacting independently still face a few connected challenges rather than one single obstacle. Financial and capital markets are regulated and deeply orchestrated systems, in which every activity has an entity that is liable for it and one that bears its consequences: the consumer, the business, the bank or the infrastructure that settles the transaction. The key considerations therefore center on how responsibility and risk are allocated: Standards are still fragmenting and interoperability remains limited. Regulation is advancing through frameworks rather than laws, and liability for an agent’s mistakes is still open. Data sits fragmented across legacy systems, and privacy is the leading AI risk for 74% of industry players and 80% of regulators, according to the World Economic Forum. Identity, security against manipulation, the systemic risk of many agents acting on the same signals, and customer trust complete the picture.
Blockchain addresses a meaningful part of this at the base layer, with a tamper-evident record of every action and programmable assets that agents can hold and move within rules written into the account. On the XRP Ledger, a protocol that was designed from inception to support financial services use cases, transactions settle in three to five seconds at a fraction of a cent, and most controls sit in the protocol rather than in a smart contract written for each agent: escrow for conditional payments, multi-signing for shared control, and deposit authorisation to limit counterparties. An agent can operate its own account, funded within its mandate, while the organisation works with infrastructure that enforces a large part of its compliance and risk controls automatically.
Q2. As AI agents move beyond generating content toward executing financial actions, which real-world applications do you expect to adopt agentic payments first?
The applications adopting agentic finance first are those where the rules are clear, the value of each transaction is contained and accountability is easy to trace.
Commerce is already live: agents complete purchases for consumers, with the card networks supplying the trust layer underneath. They register agent identities, issue tokenised credentials and authenticate what the customer approved; rules on who pays when an agent makes a mistake are only beginning to take shape. Pay-per-request access to APIs, data, tools and compute are already one of the active segments. Agent-to-machine lets agents transact within defined spend authority and controls, with payment assurance and multi-rail settlement across cards, accounts and stablecoins. Escrow-backed work follows the same logic, with funds released only once a task is delivered and verified.
Risk and compliance is one of the strongest fits. Agents assemble evidence for anti-money-laundering investigations, refresh KYC files and check activity against policy before it executes; in operations, they repair payments and reconcile trades while a person keeps release authority.
Closer to the balance sheet sit invoice payments, account reconciliation, treasury operations and autonomous B2B procurement: routine, rules-based functions where the logic is well defined and the volume is high. From there, it extends to capital allocation, such as moving idle stablecoin balances into yield vaults within a defined mandate. A distinct category is enterprise intelligence infrastructure, which connects agents to an organisation’s own data and systems under its existing permissions. The closer an agent moves to the balance sheet, the higher the bar for authorisation, oversight and accountability, and that is where the real work of adoption now sits.
Q3. How can organizations grant AI agents meaningful financial autonomy while maintaining strict human controls over permissions and spending limits?
Granting agents financial autonomy is indeed a significant risk for institutions to manage, and the market is addressing it by defining a growing set of controls, mandates and standards. The starting point is knowing who the agent is and what it has been authorised to do. Agents are increasingly registered and tied to an accountable entity, and they act under a clear mandate that sets the scope of their authority, the limits that apply and the point at which a person must step in.
Those limits are enforced by the infrastructure around the agent rather than left to the agent itself, and every proposed action should be checked against the institution’s policies before it executes. Oversight is calibrated to risk: routine, low-value actions can potentially run autonomously, while larger or irreversible ones call for human approval. Continuous monitoring, the ability to revoke access instantly and a full audit trail round this out, and supervisors are increasingly setting expectations in each of these areas. The underlying principle is a familiar one in finance: an agent should hold only the authority its task requires, for only as long as it needs it.
On the XRP Ledger, several of these controls are native: multi-signing, deposit authorization to accept funds only from approved parties, and escrow to release funds only when a condition is met. Organisations that build these layers now will ideally be able to extend autonomy safely as the technology matures.
Q4. How could emerging payment protocols like x402 change machine-to-machine commerce, particularly when it comes to accessing APIs and purchasing data?
x402 revives a status code the web reserved decades ago, HTTP 402 Payment Required, and turns it into a working payment flow. A service responds to a request with its price, the agent resends the request with a signed payment attached, and once the payment is verified and settled, the data or service is delivered. For the agent, there is no account to create, no API key to provision and no monthly invoice.
This changes the business model of the machine economy. Instead of a subscription negotiated in advance by a human, an agent buys exactly what it needs, when it needs it, from any provider that supports the standard. The next step is trust: providers will want to know which agent they are serving and on whose authority, which is where verifiable intent and agent identity meet the payment itself.
On the XRPL, x402 payments settle in XRP or RLUSD, and the XRPL AI Starter Kit connects agents to x402 and to the Machine Payments Protocol, an open standard that adds sessions for streams of small payments. Open standards are what make this scale: the more agents and providers speak the same protocol, the deeper this market becomes.
Q5. What role will stablecoins like RLUSD play in an AI-driven economy, and do you expect autonomous agents to become significant users of digital currencies?
I believe they will, and the reason goes beyond payments. A blockchain is, at its core, a shared database that makes it possible to tokenise assets that until recently could not be held, moved or traded this way, with government bonds, money market funds, invoices and commodities among them. As that list grows, so does the portfolio agents can run, rebalance and sell within a defined mandate.
Stablecoins are a core pillar of that picture because they are the bridge. Every trade, payment or treasury balance an agent handles needs a settlement leg whose value is stable and which works around the clock. Regulated, dollar-backed stablecoins such as RLUSD are designed for these price-stable flows, including invoice settlement, payroll and agent-to-agent commerce, and already settle agent payments on the XRP Ledger. With the leading regulatory frameworks such as GENIUS Act and MiCA, institutions have clearer rules on which stablecoins they can hold; the open questions sit around custody and authority, meaning who controls the keys and which assets an agent may touch.
Q6. What inspired XRPL Commons to make Agentic Finance a central theme of the October 24–25 hackathon in New York, and what key problems do you hope participants solve?
I believe agentic finance represents the next era of the digital economy. Banks, payment networks, FinTechs and technology companies are now mapping the AI-based applications that matter most to them and their clients.
In this upcoming XRPL NY Hackathon, together with our partners at Ripple, AWS, Mastercard, Ledger and QuickNode, we are expanding the XRPL ecosystem by bringing together assets, tools, capabilities and know-how to open new possibilities in on-chain finance. The hackathon connects XRPL’s native building blocks with partner expertise across its themes, including agent-to-machine payments and yield strategies. With each partner we are defining the most relevant use cases, and connecting their focus areas, tools and expertise in one place which gives builders a sandbox that would be very hard to assemble alone.
The New York hackathon is also open to existing projects that want to use it as a sprint to build out their product on the XRP Ledger. A hackathon can prove that an idea works. Building it into a product that users and institutions will trust takes sustained effort beyond the event. That is where XRPL Commons continues working closely with the strongest teams, offering technical mentorship, partner introductions and access to XRPL startup programmes.
