Ethereum Foundation Launches zkAPI to Pay for AI Models Without Revealing Identity

Foundation brings anonymous AI payments to Ethereum mainnet

The Ethereum Foundation announced zkAPI on October 1, 2026, a system that lets users pay for AI models and other metered APIs without revealing who they are. The Foundation built the system alongside the Open Anonymity (OA) Project, an open-source effort focused on AI privacy, and it is already live on Ethereum mainnet (the primary Ethereum blockchain where real transactions settle).

According to the Foundation's blog post, every AI API call today ties together an identity trail — your API key links to an account, the account links to a payment method, and the provider can connect years of your usage into a single profile. zkAPI aims to break that chain.

Key details

• Users deposit tokens like ETH or USDC (a dollar-pegged crypto stablecoin) into a vault contract on Ethereum, which stores their balance as a private note.
• To spend funds, software on the user's device generates a zero-knowledge proof — a mathematical way to prove something is true without revealing the underlying details — showing the request is backed by a funded note without exposing which one.
• A zkAPI server checks the proof, issues a temporary API key with a spending cap, and deducts usage from the user's private balance when that key expires.
• The protocol is described as experimental on the project's GitHub repository.
• zkAPI does not provide network anonymity. The gateway can still correlate requests from a stable IP address, and prompt content (personal details, writing style, conversation history) can also re-link sessions.

How the anonymous payment flow works

When a user sends a request, their device software creates a zero-knowledge proof that their balance covers the cost, without revealing which specific deposit funds the request. The zkAPI server validates this proof and hands back a short-lived API key capped at a set spending limit.

The user then sends their prompts directly to the AI model provider using that temporary key. Once the key expires, the server subtracts the usage from the user's private balance and publishes a serial number called a "nullifier." If someone tries to spend the same balance twice, the duplicate nullifier exposes the attempt without revealing anything else about the user.

What the Foundation's blog post says

The blog post, written by Vittorio Rivabella of the Foundation's dAI (decentralized AI) team, explains the problem it addresses: "Every AI API call today carries an identity." The post frames zkAPI as a way to decouple payment from identity so providers cannot build long-term profiles of users.

The post also notes that the zkAPI client exposes the standard OpenAI and Ollama APIs locally, meaning existing apps, editors, and chat clients can work with it by pointing them at localhost (the user's own machine).

First use cases and broader applications

AI chat and AI agents are listed as the first use case. The post also mentions potential applications including blockchain RPC queries, image and video generation, VPN bandwidth, and machine-to-machine payments between AI agents.

Designed by Buterin and Crapis

zkAPI implements "ZK API Usage Credits," a design Ethereum co-founder Vitalik Buterin and Ethereum Foundation dAI lead Davide Crapis published on the Ethereum Research forum on February 11, 2026.

Crapis leads the Foundation's dAI team, which the Foundation formed in September 2025 to improve Ethereum's capabilities as a settlement and coordination layer for AI. That same team developed ERC-8004, a standard for AI agent identity that launched on mainnet in January 2026.

Ken Liu, a Stanford computer science PhD candidate who works on the Open Anonymity Project, said zkAPI is "like a generalization of OA unlinkable inference that also abstracts payments away." He added that more infrastructure should exist where privacy and sovereignty are foundational and people are in control.

What is still unclear

The GitHub repository labels the protocol as experimental, which suggests the system is still in early development. The blog post acknowledges that zkAPI does not offer network-level anonymity and that prompt content can still be used to re-identify sessions. No timeline was provided for when the system might move beyond experimental status.

Sources

• The Block