Machine-payment infrastructure is moving from experimental design to measurable on-chain use, led by rapid growth in x402 activity on Base. Chainalysis said on Oct. 15 that the protocol had processed about 138.5 million transactions on the network by the end of the third quarter of 2026, rising sharply from 58,800 transactions in the third quarter of 2025.
The increase offers an early indication that software agents — programs designed to carry out tasks with limited human intervention — are beginning to use blockchain payments for more than isolated demonstrations. Yet the transaction growth also exposes the distance between a payment rail that can process automated transfers and a system that can safely let a machine manage money, identity checks, gas fees and real-world delivery.
Payment rails are arriving before agent safeguards
Developers have begun building several ways for agents to request and make payments. x402 adapts the HTTP 402 “Payment Required” status code for internet-native payments, allowing a service to charge a program before delivering data, computing capacity or another digital resource. A machine could, for example, pay for a data query or an API call as part of completing a larger task.
Other projects are pursuing similar models with different technical assumptions. Tempo has introduced its Machine Payments Protocol, or MPP, while a Linux Foundation working group with 40 members has been formed around related standards and tooling. BNB Chain released Agent Studio in July, offering tools intended to help developers build agent-based applications.
The race is increasingly focused on the payment entry point: the layer that controls quotes, routing and settlement. An HTTP-based standard such as x402 could become the payment request layer used across applications, while wallets and public blockchains are seeking to embed the same functionality closer to the transaction itself.
That competition matters because agent payments are likely to involve a very different transaction pattern from human commerce. A person may make a handful of high-value purchases each day. An agent could make many small payments while gathering information, paying for software services, coordinating with other programs or executing a defined operating budget.
Raw transaction totals, therefore, offer only a partial view of real demand. Automated payment systems can generate substantial on-chain activity without proving that users received a useful service or that the payments were economically sustainable. The next test is whether these rails can support repeated, controlled transactions tied to verifiable outcomes.
Identity and custody remain unresolved
The largest obstacle is that an agent cannot independently complete the checks required in regulated financial systems. It cannot open a bank account, fill out conventional documentation, complete know-your-customer procedures, or meaningfully assume legal responsibility during anti-money-laundering and counterparty screening.
That leaves projects working on frameworks where a human or organization retains the legal identity and asset ownership while software acts within narrow permissions. This model would allow an agent to execute a task without presenting the agent itself as an autonomous account holder with unrestricted access to financial services.
Custody introduces a separate risk. An agent with direct control over a private key could lose funds through a software vulnerability, a manipulated prompt, a faulty tool call or repeated execution after a failed task. A bot trying to solve a simple problem can also retry actions many times, turning a minor error into uncontrolled spending.
Practical systems will need spending limits, transaction restrictions, clear authority rules and revocable access. These controls are not merely user-interface features. They determine whether an agent can be trusted with funds without giving it the ability to drain a wallet or make payments outside its mandate.
Gas fees create another operational hurdle. Blockchain transaction fees are usually paid in each network’s native token, but an agent may be receiving a payment or buying a service in a stablecoin. Moving across several chains can require the agent to acquire and manage multiple gas assets, interrupting an otherwise automated workflow.
Sponsored transactions and gas abstraction aim to remove that burden. Under a sponsored model, the party initiating an action does not necessarily pay the network fee. Gas abstraction can also allow fees to be handled in an asset other than the chain’s native token, helping an agent complete a process without stopping to purchase gas.
Pharos targets protocol-level agent activity
Pharos Network has positioned its “Agent Native” strategy around this protocol-layer approach. The network says agents should be able to hold assets, call tools, coordinate with other agents and pay for actions as on-chain economic participants, rather than relying solely on AI features added inside consumer applications.
Pharos said its Pacific Ocean mainnet launched in April and promotes EVM and WASM compatibility, parallel execution through its SALI architecture, a modular SPN design and native x402 support. The network has cited roughly 30,000 transactions per second and sub-second finality as performance targets or capabilities.
Its near-term plan includes Sponsored Transactions, which Pharos says are already live on testnet, alongside gas abstraction or substitution tools. The intended result is a workflow where an agent can make a payment without separately sourcing native tokens at every stage.
The more difficult layer is scheduled later. Pharos has placed “Agent Accounts & KYA” on its 2027 roadmap, describing it as account abstraction combined with a permissions framework. The proposal follows the prevailing model for the sector: humans retain control over assets, while agents receive restricted authority to execute specified actions.
The network is also developing an Agentic Economic Activity, or AEA, framework that would assess effective actions, end-to-end delivery, continuity, security, self-healing and resistance to Sybil manipulation. Such measures would address a weakness in judging agent systems purely by transaction counts, particularly where automated programs can create large volumes of low-value activity.
Authorization is the slowest component
Current development is often divided into four layers: payment, identity, hosting and authorization. Payment infrastructure is progressing first, followed by identity tools and agent hosting. Authorization remains the least mature component because it defines the boundary between useful automation and unacceptable loss of human control.
Visa and Mastercard have both explored agent-payment concepts, but authorization models remain largely at a testing and demonstration stage. A workable system must establish who approved a payment, what an agent was permitted to do, whether a service was delivered, and who is responsible when an irreversible on-chain transfer produces an unsatisfactory off-chain result.
This is especially relevant for small, frequent payments. Blockchains can settle transfers quickly and irreversibly, while the service being bought may be delayed, incomplete or impossible to verify automatically. Reputation systems, dispute processes and performance records may become as necessary as payment protocols for unattended commerce.
The market’s likely rollout remains gradual. Payment and identity tools are being targeted for adoption through 2026 and 2027, while stronger authorization and multi-agent coordination are expected to take longer. Pharos has placed complete end-to-end workflows on a 2028 horizon.
Fragmented standards could slow that schedule. Competing protocols across chains may split liquidity and force developers to support multiple payment systems, while regulators could limit unsupervised agent trading or spending. The projects most likely to gain durable use will be those that combine low-friction settlement with identity, custody controls and clear human authorization, rather than simply producing the highest volume of automated transfers.
To see how AI-driven markets evolve alongside machine payments, explore 2026’s narrative drivers in crypto trading.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
