ETH Foundation launches zkAPI for private AI

TLDR

Ethereum Foundation and Open Anonymity Project have launched zkAPI on Ethereum mainnet, letting users pay for AI and other APIs without tying their identity to what they ask.

1. zkAPI uses a funded onchain vault and zero?knowledge proofs to issue short?lived API keys so providers can be paid without seeing who is paying.
2. The system hides the payment link, not the prompt; AI providers still see content and network metadata, and zkAPI is explicitly labeled experimental.
3. This is a major step in Ethereums AI roadmap, hinting at future private agent payments, but real impact depends on adoption, audits, and stronger privacy tooling.

Deep Dive

1. How zkAPI Works

According to the Ethereum Foundations launch post and multiple reports, zkAPI is live on Ethereum mainnet and built with the Open Anonymity Project to enable private, prepaid API access for AI and other metered services. Users deposit ETH or stablecoins like USDC into a vault contract; those deposits become notes tracked onchain but referenced privately offchain.

When a user wants to call an AI model, their device generates a zero?knowledge proof that there is enough unspent credit without revealing which deposit is theirs. A zkAPI server verifies the proof and issues a short?lived API key with a spending cap; prompts go directly to the AI provider, while usage is settled later from the vault balance, with a nullifier to prevent double spending. This design implements the ZK API Usage Credits proposal co?authored by Vitalik Buterin and Davide Crapis and now turned into a working mainnet implementation.

2. Privacy Scope And Limits

Reports stress that zkAPI provides payment unlinkability, not full anonymous AI use. Providers still receive the prompts and responses, and gateways can correlate sessions through IP addresses, timing, reused context, and writing style. Even in the most private mode, the payment server learns that a valid note spent a certain amount, while the AI provider sees the content, just not the payers billing identity.

The projects repository and documentation describe the protocol as experimental, with no comprehensive public audit yet, and suggest using tools like Tor for stronger network?level privacy. In practice, the anonymity users experience will depend on client configuration, routing choices, and how much personal detail they put in their prompts.

3. Impact And What To Watch

zkAPI fits directly into Ethereums emerging role as a settlement and coordination layer for AI, following earlier research and standards work by the Foundations dAI team. Immediate use cases include private AI chat and agents, but the same pattern could extend to blockchain RPC, image and video generation, VPN bandwidth, and machine?to?machine payments between AI agents.

For crypto users and builders, the key variables to watch are: real usage volumes through providers like OpenRouter, independent audits of the contracts and circuits, integration into wallets or agents, and whether stronger privacy layers (network anonymity, better client defaults) are added on top. If those pieces mature, zkAPI?style payment rails could become a standard way to buy AI and other services using Ethereum without handing over a long?term usage profile.

Conclusion

zkAPI turns a research design into live Ethereum infrastructure that separates payment identity from AI and API usage, a meaningful privacy upgrade compared with conventional API keys. Its launch reinforces Ethereums positioning at the intersection of cryptography and AI, but the system is early and its privacy guarantees are bounded. The real significance will emerge as adoption, audits, and surrounding privacy tooling catch up, and as more AI and infrastructure providers choose to plug into this kind of cryptographic payment layer.