Ethereum Foundation ships zkAPI, a pay-for-AI system that hides who's paying

The Ethereum Foundation on Thursday launched zkAPI, a system meant to let people pay for AI models and other metered APIs without tying the requests to their identity, according to a Foundation blog post reported by The Block. The Foundation says it is live on Ethereum mainnet.

The mechanics, as described in the Foundation's post: a user deposits tokens such as ETH into an Ethereum vault, then authorizes spending with a zero-knowledge proof. A zkAPI server verifies the proof and issues a temporary API key carrying a spending cap; the user sends prompts directly to the AI provider with that key, and when it expires the server deducts usage from the user's private balance. Each payment publishes a serial number — a "nullifier" — so that an attempt to spend the same balance twice produces a duplicate and is rejected. The Foundation says the client exposes the standard OpenAI and Ollama APIs on localhost, so existing apps point at it unchanged.

The build is a collaboration with the Open Anonymity Project, an open-source AI privacy effort. It implements "ZK API Usage Credits," a design that Foundation dAI lead Davide Crapis and Ethereum co-founder Vitalik Buterin published on the Ethereum Research forum on Feb. 11. Crapis's dAI team, formed in September 2025, earlier shipped ERC-8004, an AI-agent identity standard that went live on mainnet in January. The announcement post was written by the dAI team's Vittorio Rivabella.

The Foundation is upfront about the limits. zkAPI provides no network-level anonymity: a gateway can correlate requests from a stable IP, and the post suggests routing through Tor for stronger privacy. Sessions can also be re-linked through prompt content — personal details, writing style, conversation history. The project's GitHub repository calls the protocol experimental.

Key facts

• zkAPI launched Thursday on Ethereum mainnet, built by the Ethereum Foundation with the Open Anonymity Project (Ethereum Foundation blog, via The Block).
• Users deposit ETH or other tokens into a vault and authorize spending via zero-knowledge proof; a server issues a capped, temporary API key (EF blog).
• A "nullifier" serial number blocks double-spending of a balance (EF blog).
• Implements the "ZK API Usage Credits" design Crapis and Buterin posted to the Ethereum Research forum on Feb. 11 (The Block).
• No network anonymity; sessions can be de-anonymized via IP or prompt content; GitHub labels the protocol experimental (EF blog).

The real-world read

This is a first-party launch announcement — the Foundation describing its own system, with The Block the only outlet carrying it so far. The privacy framing ("every AI API call today carries an identity") is the Foundation's own. To its credit, the post discloses the holes rather than burying them: no network anonymity, trivial re-linking by IP or writing style, and an "experimental" tag on the code. Those caveats matter, because a payment layer that hides your wallet while your prompts and IP still give you away offers narrower protection than the headline suggests. A supportive quote came from Stanford PhD candidate Ken Liu, who works on the Open Anonymity Project — an interested party, not an outside reviewer.

Opinion, and whose

Buterin, in a February X post, framed cryptographic payment mechanisms for AI services as part of his near-term vision for Ethereum's role in AI. Liu said "more infrastructure should exist where privacy and sovereignty is foundational." Both are proponents of the work; neither claim is independently verified.

Sources

• The Block (Tier 2, reputable secondary), Oct. 2, 2026 — reported the launch, mechanics, timeline and design provenance, citing the Ethereum Foundation's blog post as its primary source. The underlying Foundation blog post is a first-party announcement of the Foundation's own product; its promotional framing is noted above and discounted accordingly.

This is news coverage, not financial or security advice.