The framework, created alongside the Open Anonymity Project, brings to life an architecture detailed in February by Ethereum co-creator Vitalik Buterin and Ethereum Foundation dAI Lead Davide Crapis.
As outlined in a blog post released on Thursday, the Ethereum Foundation has rolled out zkAPI, a protocol enabling individuals to pay for artificial intelligence models and other metered APIs anonymously.
The Foundation collaborated with the Open Anonymity (OA) Project—an open-source initiative focused on artificial intelligence privacy—to develop zkAPI, which is now operational on the Ethereum mainnet, as noted in their announcement.
“Today, every interaction with an AI API comes with an identity,” the publication explained. “Your API key links to a specific profile, that profile ties to a payment source, and every query you submit links directly to the historical logs of both. Consequently, service providers can stitch years of your activity together into a single comprehensive file.”
Ken Liu, a PhD student in computer science at Stanford and contributor to the Open Anonymity Project, remarked in an afternoon post on X that “It represents a broader concept of OA unlinkable inference that simultaneously abstracts away payment mechanics.” He further commented that “we need more infrastructure where privacy and user sovereignty serve as foundational pillars, putting individuals back in control.”
How zkAPI works
According to the update, engaging with zkAPI requires users to first deposit assets like ETH or USDC into an Ethereum vault contract, which translates the deposited amount into a private balance note. When spending these funds, client software running on the user’s local machine constructs a zero-knowledge proof verifying that the transaction is backed by a valid, funded note without revealing the identity of the specific deposit.
Following this, a zkAPI validator examines the zero-knowledge proof and provides a short-lived API key constrained by a spending limit. The user then submits their queries directly to the AI service provider alongside this temporary key, and once the session ends, the server debits the appropriate amount from the user’s private credit pool.
Additionally, every time a payment for AI services is made, a distinct serial identifier referred to as a “nullifier” gets broadcasted. “Attempting to utilize the same funds twice generates a duplicate nullifier, exposing the double-spending attempt while keeping all other information secure,” the document emphasizes.
The client software integrates seamlessly with pre-existing AI utilities, per the announcement. “Your local machine hosts the client, which mirrors standard OpenAI and Ollama APIs,” the text outlines. “Pre-existing applications, text editors, and chat interfaces can connect instantly by directing traffic to localhost.”
Use cases and limitations
The write-up highlights conversational AI and autonomous agents as the initial primary applications. Additional potential uses mentioned include blockchain RPC requests, image and video generation, bandwidth for virtual private networks, and automated machine-to-machine transactions executed by AI agents.
Nevertheless, the zkAPI framework does not guarantee network-level anonymity, the post warns. The gateway server could theoretically link requests originating from a static IP address, meaning privacy-conscious users might prefer routing their internet traffic through Tor—a specialized anonymity network that bounces packets across multiple nodes, frequently utilized by privacy advocates—according to the article.
Furthermore, active sessions might still be deanonymized through the semantic content of prompts, including personal details, unique writing styles, or past dialogue history. The public GitHub repository for the zkAPI initiative explicitly labels the software protocol as experimental.
Design co-authored by Vitalik Buterin
The zkAPI mechanism puts into practice “ZK API Usage Credits,” a conceptual framework originally shared on the Ethereum Research forum by Davide Crapis and Ethereum co-founder Vitalik Buterin on February 11.
Crapis heads the decentralized AI (dAI) division within the Ethereum Foundation, which was established in September 2025 to enhance Ethereum’s utility as a settlement and coordination foundation for the burgeoning artificial intelligence economy. This same division also engineered ERC-8004, a decentralized standard for AI agent identities that debuted on the mainnet during January.
The official announcement detailing zkAPI was authored by Vittorio Rivabella, a member of the dAI division.
Earlier in February, Buterin emphasized cryptographic payment systems tailored for AI utilities as an essential component of his near-term outlook regarding Ethereum’s strategic role within an AI-dominated ecosystem, as expressed in an X post.
Originally published at https://www.theblock.co/news/defi/2026-10-01-ethereum-foundation-launches-zkapi-417504.
