"Every AI API call today carries an identity." That's how the Ethereum Foundation described the current state of AI service payments in a blog post—API keys are tied to accounts, accounts are tied to payment methods, and every single prompt gets folded into the same record. This is exactly the problem zkAPI is built to solve.
Developed jointly by the Ethereum Foundation and the open-source AI privacy project Open Anonymity (OA) Project, zkAPI officially went live on Ethereum mainnet on October 1 (Eastern Time). The system lets users pay for AI models and other usage-based API services without letting the service provider or payment server know who's paying.
Ken Liu, a Stanford computer science PhD student and OA Project member, described zkAPI in an X post the same afternoon as "a generalization of OA's unlinkable inference, with payment abstracted away alongside it." He added that "there should be more infrastructure that treats privacy and sovereignty as foundational, giving people real control over themselves."
Vaults, Private Notes, and One-Time Nullifiers
Before using zkAPI, users must first deposit ETH or USDC into a vault contract, and this deposit gets recorded as a "private note." When it's time to spend those funds, software on the user's device generates a zero-knowledge proof confirming the request is backed by real funds—without revealing which specific deposit it came from.
Once the zkAPI server verifies this proof, it issues a temporary API key with a spending cap and time limit. The user then sends their prompts directly to the AI model provider along with this key. After the key expires, the server deducts the corresponding usage from the user's private balance.
Each payment simultaneously publishes a serial number called a "nullifier." According to the official explanation, if someone tries to spend the same balance twice, it produces a duplicate nullifier—"this exposes the attempt itself, but nothing else." The team also noted that the client exposes standard OpenAI and Ollama API interfaces locally on the user's machine, so existing apps, editors, and chat clients can work right out of the box just by pointing to localhost.
AI Chat Is Just the First Step—But It's Not Fully Anonymous
The first use case listed in the official blog post is AI chat and agents, with other potential applications including blockchain RPC queries, image and video generation, VPN bandwidth, and machine-to-machine payments between AI agents.
That said, zkAPI doesn't provide network-level anonymity. The team openly admits that the gateway could still link different requests through a fixed IP address, so users wanting stronger privacy can route their traffic through anonymity networks like Tor. Beyond that, the conversation content itself—personal information, writing style, or chat history—could also allow different sessions to be linked back to the same person. zkAPI's GitHub project page also labels the protocol as "experimental."
zkAPI implements the design called "ZK API Usage Credits," which Davide Crapis, head of the Ethereum Foundation's dAI team, and Ethereum co-founder Vitalik Buterin published on the Ethereum Research forum on February 11. The dAI team was formed in September 2025 with the mission of strengthening Ethereum's role as a settlement and coordination layer for AI—it previously developed ERC-8004, a standard for AI agent identity that launched on mainnet back in January. The blog post announcing zkAPI's launch was written by Vittorio Rivabella of the dAI team. Buterin himself listed encrypted payment mechanisms designed for AI services as one of his visions for Ethereum's near-term role in AI, in an X post back in February.
