All news is rigorously fact-checked and reviewed by leading blockchain experts and seasoned industry insiders.
- zkAPI is now live on Ethereum mainnet.
- Payments cannot be directly linked to individual AI sessions.
- The system uses private prepaid credits instead of persistent billing identities.
- Glamsterdam reaches Sepolia on Oct. 6.
The Ethereum Foundation and Open Anonymity Project have launched zkAPI on Ethereum mainnet, giving users a way to pay for AI inference without automatically linking individual requests to the wallet that funded them.
The protocol turns prepaid credits into private notes and uses zero-knowledge proofs to authorize spending. An AI provider can process a request while remaining separated from the original billing identity, addressing a privacy problem created when API keys, payment methods and long-running prompt histories belong to the same account.
zkAPI implements a design published in February by Ethereum Foundation researcher Davide Crapis and Ethereum co-founder Vitalik Buterin. The Open Anonymity Project subsequently developed the client, server and smart contracts now running on mainnet.
One Deposit Can Fund Multiple Private AI Sessions
zkAPI avoids both conventional account-based billing and the need to make an onchain payment for every API call.
A user first deposits credits into an Ethereum vault. The balance becomes a private note represented through a cryptographic commitment. When the user wants API access, software running locally generates a proof that a valid note has sufficient funds without identifying the note or its original deposit.
The payment flow is relatively compact:
- Deposit: Funds enter an Ethereum smart contract.
- Prove: The local client proves that sufficient private credit exists.
- Authorize: A short-lived API key receives a defined spending allowance.
- Settle: Actual usage is charged against the private balance after the session.
The live implementation uses Groth16 proofs over BN254, Poseidon hashing and a 32-level Merkle tree. Nullifiers prevent the same balance from being spent twice without revealing which deposit funded a legitimate request.
The current default mainnet deployment uses real USDC for API credits, with ETH required for transaction gas.
The broader design can support other billing tokens.
zkAPI Solves a Different Problem From x402
The launch also adds another model to the emerging market for machine-native API payments.
Coinbase’s x402 protocol removes conventional accounts from API payments by using the HTTP 402 “Payment Required” mechanism. An API can return a price, the client signs a stablecoin payment and the request proceeds once that payment is verified. Coinbase already uses x402 for services where AI agents can pay per query without maintaining a traditional account.
zkAPI targets a different layer of the problem.
Its defining feature is not simply programmable payment, but unlinkability between funding and usage. Instead of attaching a payment authorization to each request, a user deposits once and later proves access to prepaid credit without exposing which deposit is being spent.
That distinction could matter for AI because prompts can contain substantially more personal information than an ordinary digital purchase. The financial trail itself becomes another identifier when every request belongs to the same billing account.
Payment Privacy Still Does Not Mean Anonymous AI
zkAPI deliberately has a narrower privacy boundary than fully anonymous inference.
In its runtime-key mode, the zkAPI payment server does not receive prompts. They travel directly from the user’s device to the AI provider using a short-lived key. The provider, however, still sees the prompts and responses because it performs the inference.
It can also observe network information such as an IP address and potentially correlate sessions through timing, reused conversation histories, personal details or writing patterns.
The Ethereum Foundation therefore separates payment privacy, network anonymity and content privacy. Tor or similar tools address network exposure, while confidential computing or local models can tackle different aspects of prompt confidentiality. zkAPI itself primarily removes the billing link.
That boundary is important because the protocol does not make AI usage invisible. It prevents the payment system from automatically becoming a persistent identity layer for that usage.
Glamsterdam Heads to Sepolia Days After zkAPI Launch
The mainnet deployment arrives just before another Ethereum infrastructure milestone.
The Glamsterdam network upgrade is scheduled to activate on Sepolia on Oct. 6 at 13:53:36 UTC, at epoch 353,024. Hoodi and Ethereum mainnet activation dates have not yet been decided.
Glamsterdam is separate from zkAPI, but its changes affect the Layer 1 environment on which applications and settlement systems operate. The upgrade introduces enshrined proposer-builder separation, block-level access lists and changes to gas pricing intended to better reflect execution and state-growth costs.
Block-level access lists are particularly relevant to Ethereum’s broader scaling direction because identifying state dependencies in advance can support more parallelized execution. Glamsterdam also continues work on increasing Layer 1 capacity following Fusaka.
It would be premature, however, to claim that Glamsterdam will directly make zkAPI cheaper. zkAPI verifies request proofs off-chain, while Ethereum is used for the vault and settlement components. Any cost improvement for the protocol will depend on actual post-upgrade network conditions and how its onchain operations are used.
For Sepolia users, the Oct. 6 activation is primarily an infrastructure event. Node operators need compatible execution and consensus clients, while regular ETH holders do not need to take action for the testnet fork.
Private Credits Could Extend Beyond AI
AI inference is zkAPI’s first use case, but its architecture is not tied to language models.
The Ethereum Foundation identifies blockchain RPC calls, image and video generation, VPN bandwidth and machine-to-machine services as other metered applications where a customer may want to prove payment without creating a persistent financial identity. Ethereum Foundation Blog
That gives zkAPI a different adoption test from a consumer privacy application. Its longer-term relevance will depend on whether API providers are willing to support private usage credits alongside conventional accounts and newer payment standards such as x402.
With the protocol now live and Glamsterdam moving onto Sepolia days later, Ethereum is testing changes at two different layers: how applications can privately pay for services, and how the underlying network executes and prices the activity that eventually reaches Layer 1.
