Ethereum Foundation Launches zkAPI: Privacy Payments for AI

The Ethereum Foundation went live on mainnet on 2 October 2026 with zkAPI, a payment protocol that lets users purchase access to AI models and other metered API services without exposing who they are. Built jointly with the Open Anonymity Project, it is the first production deployment of a design co-authored by Ethereum co-founder Vitalik Buterin and Ethereum Foundation dAI lead Davide Crapis, originally published in February 2026. For accounting firms, CFOs, and auditors already grappling with the market structure developments reshaping digital asset accounting software, zkAPI introduces a new category of on-chain activity that demands immediate classification attention.

What zkAPI Actually Does

At its core, zkAPI is a privacy-preserving payment rail layered on top of Ethereum. The system breaks the usual link between a payer's identity and their API usage, something that is standard practice in virtually every commercial AI service today.

The deposit and proof mechanism

A user begins by depositing tokens, ETH or USDC, into a vault smart contract on Ethereum mainnet. The vault records the user's balance as a private "note" rather than a publicly labelled account balance. When the user wants to pay for an AI query or other API call, client software running locally on their device generates a zero-knowledge proof. That proof cryptographically demonstrates that the request is backed by a funded note, without revealing which deposit, which wallet, or which prior transaction funded it.

A zkAPI server validates the proof and responds by issuing a temporary API key with a defined spending cap. The user sends their prompts directly to the AI model provider using that key. When the key expires, the server deducts the consumed usage from the private balance. A "nullifier," a unique serial number, is published on-chain for each payment event. If a user attempts to spend the same balance twice, a duplicate nullifier appears and the double-spend attempt is exposed, though the payer's identity is still not revealed in that process.

Where the anonymity ends

The Ethereum Foundation is clear about the protocol's limits. zkAPI does not provide network-level anonymity. A gateway can potentially correlate multiple requests originating from a stable IP address. The content of prompts, writing style, personal details, or a continuing conversation thread can also allow sessions to be re-linked. Users who want stronger network privacy can route traffic through Tor, though the Foundation describes the protocol overall as experimental at this stage.

Intended use cases

The Foundation's launch post lists AI chat and autonomous agents as the primary targets. Additional use cases include blockchain RPC queries, image and video generation services, VPN bandwidth purchasing, and machine-to-machine payments between AI agents. The client is designed to expose standard OpenAI and Ollama APIs on a user's local machine, meaning existing tools, editors, and chat clients can work with zkAPI by pointing at localhost with no further integration work.

The Research Lineage and Ethereum's dAI Strategy

zkAPI is not an isolated experiment. It is one piece of a broader Ethereum Foundation initiative to position Ethereum as a settlement and coordination layer for AI-related services. Crapis's dAI team also developed ERC-8004, a proposed standard for AI agent identity on Ethereum that appeared in January 2026. Buterin highlighted cryptographic payment mechanisms for AI services as a strategic priority in a February 2026 post on X. The zkAPI launch therefore represents a deliberate progression from research to mainnet deployment within roughly eight months, which is a fast timeline by Ethereum standards.

For accountants tracking the Ethereum ecosystem, this matters because the dAI team's work is establishing new classes of on-chain transactions, private vault deposits, ZK-proof-authorised disbursements, nullifier publications, that have no direct precedent in existing chart-of-accounts templates or bookkeeping workflows.

Accounting Classification: Where Does a zkAPI Deposit Sit?

The first practical question for any finance team whose clients, or whose own organisation, uses zkAPI is how to classify the vault deposit. Several frameworks are relevant depending on jurisdiction and applicable standard.

Under IFRS and US GAAP digital asset guidance

Under IFRS, digital assets are most commonly held as intangible assets under IAS 38, or as inventory under IAS 2 if held for sale in the ordinary course of business. A zkAPI vault deposit does not sit neatly in either bucket. The deposited ETH or USDC is not held for appreciation or sale. It is prepaid consideration for a future service, placing it closer to a prepayment or advance payment asset on the balance sheet. The precise treatment will depend on whether the deposited amount is refundable and on the expected consumption timeline.

Under US GAAP, following the FASB's updated guidance on digital assets that introduced fair-value measurement for certain crypto holdings, the classification of vault deposits may be equally ambiguous. If the deposited token is USDC and FASB's proposed stablecoin-as-cash-equivalent treatment advances, the deposit could qualify as a cash equivalent, provided the relevant conditions on redeemability are met. If the deposited asset is ETH, fair-value measurement through net income applies, and any change in ETH value between the deposit date and the date of consumption would generate a recognised gain or loss.

Practical bookkeeping steps

Finance teams using any crypto bookkeeping software should establish a separate ledger code for zkAPI vault deposits at the point of transfer. The deposit represents an asset, either a prepayment or a digital asset held at fair value, depending on the token type and applicable standard. Each nullifier-confirmed payment event should trigger a derecognition of the corresponding portion of the vault balance and recognition of an expense for the AI service consumed. Because nullifiers are published on-chain, they provide a verifiable, timestamped record of each spend event, which is useful for audit trail purposes even if the underlying payer identity is obscured.

AML and Compliance Implications

This is where zkAPI becomes genuinely challenging for compliance officers and the firms advising them. The protocol is specifically engineered to sever the on-chain link between a payer and a payment. That design objective, while legitimate from a data-privacy standpoint, intersects directly with transaction-monitoring obligations under the Financial Action Task Force's Recommendation 16 (the Travel Rule) and equivalent national frameworks.

Travel Rule considerations

The Travel Rule requires virtual asset service providers to collect and transmit originator and beneficiary information for transfers above threshold values. A vault deposit into a zkAPI contract is an on-chain transfer to a smart contract. Whether it constitutes a "transfer" for Travel Rule purposes depends on how the relevant national authority classifies interactions with non-custodial smart contracts. Several jurisdictions, including the EU under MiCA's implementing rules, are still developing their position on this point. Understanding how MiCA enforcement gaps affect privacy-adjacent crypto services is directly relevant here, as zkAPI's privacy design would be subject to scrutiny under any regime that requires attributable transaction data.

Risk-based monitoring in practice

Firms whose clients interact with zkAPI vaults should document the purpose of those deposits explicitly in their AML files. The published nullifier record does confirm that a spend event occurred, and a responsible compliance framework should capture each nullifier event alongside the notional fiat value of the consumed service at the time of the transaction. This does not resolve the identity gap that the ZK proof deliberately creates, but it does produce an auditable ledger of usage amounts and timestamps.

For corporate treasury teams running autonomous AI agents that pay for their own compute costs via zkAPI, the compliance picture is more complex still. Machine-to-machine payments between AI agents, one of the explicitly listed use cases, would in principle generate a stream of micro-transactions with no human-visible originator identity. Internal policy should define who in the organisation is responsible for the vault top-up transactions, treating those deposits as the attributable point of control even if the downstream agent payments are pseudonymous.

Implications for Digital Asset Accounting Software Workflows

Most enterprise-grade digital asset accounting software currently classifies inbound and outbound transfers based on wallet address matching and exchange API feeds. zkAPI vault deposits will appear on-chain as transfers from a known wallet to a smart contract address. The deductions, the individual spend events, will not appear as conventional transfers back out. Instead, the vault's internal state changes, confirmed by nullifier publication, drive the accounting entries.

This means standard wallet-import workflows will likely misclassify zkAPI activity unless rule sets are updated. Finance teams should work with their ethereum accounting processes to add specific recognition rules: identify the vault contract address, treat inbound transfers to it as prepayment assets, and treat nullifier-confirmed spend events as expense recognition, with fair-value adjustment where the deposited asset is ETH rather than a stablecoin.

Audit teams should also consider how they will obtain sufficient appropriate evidence for zkAPI balances. The vault balance exists as a private note in the smart contract's state. While the blockchain's public state confirms the contract's aggregate holdings, confirming an individual client's specific unspent balance may require the client to generate a ZK proof of their own balance, a novel audit procedure with no established standard as yet.

What Finance Teams Should Do Now

zkAPI is live on mainnet today, not a testnet experiment. Given that its use cases include enterprise-relevant scenarios such as corporate AI agent payments and RPC query costs, finance teams cannot defer these questions to a later review cycle.

Immediate actions

First, update your digital asset accounting software rule sets to recognise the zkAPI vault contract address and classify inbound transfers correctly as prepayments or digital asset holdings, depending on the deposited token. Second, establish an internal policy that designates a named individual or team as responsible for vault top-up transactions, creating the attributable originator record that AML frameworks require even when downstream spend events are pseudonymous. Third, engage your auditors now about what evidence standard they will apply to unspent vault balances, before those balances appear in year-end financials. Fourth, monitor regulatory guidance in your key jurisdictions on smart contract interactions under the Travel Rule. The EU, UK, and US are all at different stages of that analysis, and a position that is compliant today may require adjustment as guidance firms up.

For accounting firms advising crypto-active clients, adding zkAPI as a specific line item in client onboarding questionnaires is a low-cost step that could prevent a significant classification error downstream. If a client is depositing material amounts of ETH or USDC into vault contracts to fund AI agent operations, you want to know before the year-end ledger review, not during it.

Frequently Asked Questions

Is a zkAPI vault deposit a capital outflow or an operating expense?

Neither, at the point of deposit. The deposit is best treated as a prepayment asset because no service has yet been consumed. The expense is recognised progressively as nullifier events confirm service consumption. The deposited token's fair-value movements between deposit and consumption also generate gains or losses if the asset is ETH rather than a stablecoin.

Does zkAPI activity trigger Travel Rule obligations?

This depends on the jurisdiction and on whether the regulator treats a transfer to a non-custodial smart contract as a covered transaction. No major authority has issued specific guidance on zkAPI or similar ZK vault protocols as of October 2026. Firms should apply a precautionary approach, document the purpose and originator of vault top-up transactions, and monitor developments under MiCA's implementing rules and FinCEN's evolving DeFi guidance.

How should auditors verify an unspent zkAPI vault balance?

Standard blockchain confirmation procedures will confirm the aggregate holdings of the vault contract, but not an individual client's specific balance. Verifying a client's balance may require the client to generate a ZK proof of their note. No IAASB or PCAOB standard addresses this procedure yet, so auditors will need to document their approach in the audit file as a significant judgement.

How are AI service costs paid via zkAPI recognised in the income statement?

Each nullifier-confirmed spend event represents consumption of the prepaid service. The corresponding portion of the vault balance should be derecognised and an operating expense recognised at that point, categorised according to the nature of the AI service. For AI agent compute costs, this would typically fall within technology or research and development expense depending on the activity.

What are the VAT or indirect tax implications of zkAPI payments for AI services?

The supply of AI API access is generally a digital service for VAT/GST purposes. The pseudonymous nature of zkAPI payments does not alter the underlying tax character of the supply. However, establishing the place of supply and the customer's status for VAT reverse-charge purposes requires knowing the customer's location and business status, information that the ZK design deliberately obscures. Firms should ensure that any use of zkAPI for taxable AI service purchases is accompanied by separate documentation of the customer's jurisdiction and VAT registration status to support the correct VAT treatment.

Source: The Block