Ethereum’s mainnet now hosts a system designed to let people pay for AI services without connecting their identity to the questions they ask. The Ethereum Foundation and the Open Anonymity Project announced zkAPI on October 1, describing a model in which users prepay with assets such as USDC and receive temporary, spending-limited API keys through zero-knowledge proofs.
The launch addresses a basic tension in cloud-based AI: the service provider needs to know that a request has been paid for, while users may not want their payment identity attached to sensitive prompts. According to Decrypt, zkAPI is intended to separate those two functions rather than make the AI provider responsible for both.
How zkAPI separates payment from prompts
A user first deposits credits into a vault contract on Ethereum. The Foundation says the deposit can use ETH, USDC and other supported assets. After that transaction, the user’s balance is represented as a private note that can be spent by the user without being publicly connected to the original deposit.
When the user wants to access an AI service, software running on the user’s device creates a zero-knowledge proof. In general terms, a zero-knowledge proof allows one party to demonstrate that a statement is true without disclosing the underlying information. In zkAPI’s process, the proof shows that the user has enough available funds and has not already spent them, without revealing the user’s identity or payment history to the AI provider.
A server verifies the proof and returns a short-lived API key. The key functions like a password for software communicating with an AI service, but it is restricted by a spending cap and remains only in the device’s memory. The user’s requests are then sent to the AI provider with the temporary key, rather than with a name or payment details.
Once the key expires, the provider supplies a signed record of usage. The user is charged for the amount actually consumed, not the full spending limit attached to the key. The intended division of information is central to the design: the AI provider can see the prompt, while the payment system can see the expenditure, but neither side is supposed to connect the two.
OpenRouter, OA Chat and the project’s status
The implementation was developed with the Open Anonymity Project. The launch also includes OA Chat, a browser-based private chatbot that does not require installation. The project’s code is available on GitHub.
The documentation identifies OpenRouter as the AI provider behind the keys. OpenRouter provides access to hundreds of AI models through a single connection, and applications that support the OpenAI format can connect by directing requests to a local address.
The protocol remains experimental. Its repository does not list a formal audit, an important qualification for a system handling payments and access to AI services. The material supplied for the launch also says providers would need to support proof-based authorization and settle signed usage receipts if they wanted to integrate directly with zkAPI.
The Foundation framed the project around the privacy of AI conversations. People may use AI tools for questions involving health, finances or personal concerns, creating a record that can reveal more than an ordinary payment might. The source material also points to a May 2025 federal court order requiring OpenAI to preserve output logs, including chats users had deleted, in a copyright case brought by The New York Times and other publishers. That episode illustrates why the storage and control of AI conversations has become a practical privacy issue.
Ethereum’s broader AI and machine-payment plans
The technical direction behind zkAPI follows a February 2026 proposal on Ethereum Research by Davide Crapis and Vitalik Buterin. Crapis leads the Ethereum Foundation’s dAI team, which aims to position Ethereum as a settlement and coordination layer for AI systems. Buterin’s February 10 roadmap included zero-knowledge payments for private API use.
The Foundation’s work also fits a wider effort to support transactions between software agents. In August 2025, Crapis and Marco De Rossi proposed ERC-8004, a standard intended to help autonomous AI agents discover one another, verify identity and transact. Crapis predicted that machines could account for most Ethereum traffic within three to five years, although that is a forecast rather than a current measurement.
The same payment structure could potentially be used for other metered digital services, including blockchain data queries, image and video generation, VPNs and machine-to-machine payments. Direct adoption would depend on providers accepting proofs in place of conventional API keys and processing signed usage records.
zkAPI is aimed at users who still need cloud-based AI but want the billing layer separated from their identity. It complements, rather than replaces, local AI models: in April, Buterin demonstrated running an open-source model on his own Nvidia 5090 graphics card, a setup in which data remains on the user’s device. zkAPI addresses the different case where prompts are sent to outside servers.
Market picture
Moneyiar data at 09:23-09:24 on October 3, 2026, showed Ethereum at $2,684.5962209773, down 2.36% over 24 hours and up 0.04% over seven days. USDC was priced at $0.9999868313, with a 0.02% 24-hour gain and a 0.01% seven-day gain.
Elsewhere, Shiba Inu stood at $0.0000056556, down 5.02% in 24 hours and 3.42% over seven days. Uniswap traded at $9.2394725104, up 2.05% over 24 hours but down 4.19% over seven days.
