The Ethereum Foundation and the Open Anonymity Project launched zkAPI on Ethereum mainnet on October 1, a protocol that lets users prepay for AI services with USDC or ETH and receive capped, short-lived API keys verified through zero-knowledge proofs.

The design implements a February 2026 Ethereum Research proposal co-authored by Davide Crapis, who leads the Foundation's dAI team, and Vitalik Buterin. The dAI group's stated mandate is positioning Ethereum as the settlement and coordination layer for machine-driven commerce.

"Prompts are personal. People ask AI models about their health, their finances, their doubts. Under the current model, using AI means handing a running transcript of your thinking to whoever holds the billing relationship," the Foundation argued in its announcement. "zkAPI separates payment from identity."

The mechanics work as follows. A user deposits credits into a vault contract on Ethereum in a single ordinary transaction. The balance then exists as a private note — digital cash that, according to the Foundation, only the holder can spend and that nobody can trace back to the deposit. For each query, client-side software generates a zero-knowledge proof confirming sufficient unspent funds without revealing the balance itself. The server verifies the proof and returns a temporary API key capped at a dollar amount and held only in device memory. When the key expires, the provider issues a signed usage receipt and the user is charged for actual consumption, not the cap. The AI provider sees the request; the payment layer sees the spend; neither can link the two.

The launch comes with OA Chat, a browser-based private chatbot requiring no installation, and the code is open on GitHub. Project documentation identifies OpenRouter, the aggregator providing access to hundreds of AI models through a single connection, as the AI provider behind the keys. Any application speaking the OpenAI API format can integrate by pointing at a local address.

The privacy gap the protocol addresses is not hypothetical. In May 2025, a federal court ordered OpenAI to preserve its output logs — including chats users had deleted — as part of the copyright lawsuit brought by The New York Times and other publishers. That ruling demonstrated that prompt logs can become discoverable evidence regardless of user expectations.

zkAPI also extends a broader Ethereum-AI agenda that Crapis and Buterin have developed over the past year. In August 2025, Crapis and Marco De Rossi proposed ERC-8004, a token standard defining how autonomous AI agents discover each other, verify identity and transact. Crapis has predicted that machine traffic will constitute most Ethereum activity within three to five years. Buterin's February 10 roadmap explicitly included zero-knowledge payments for private API use as a component, and in April he demonstrated the complementary approach of running open-source models locally on an Nvidia 5090. zkAPI covers the remaining case: prompts that still travel to third-party servers.

The Foundation's post suggests the same plumbing could meter other services — blockchain data queries, image and video generation, VPNs, and machine-to-machine payments. Direct integration would require providers to accept a zero-knowledge proof in place of a conventional API key and settle against signed usage receipts.

The repository labels the protocol experimental and does not list a formal audit, meaning institutional adoption will likely wait on security review of the vault contracts and proof system.