ERC-8028, known as the Data Anchoring Token (DAT) standard, is a proposed Ethereum token specification that represents AI datasets, models, and inference outputs as semi-fungible assets with built-in usage quotas and revenue-share routing. It was submitted to the Ethereum Improvement Proposal process in 2025 by the LazAI team as pull request #1219 on the ethereum/ERCs GitHub repository, where it remains in draft status.
​

The proposal arrived alongside a wider 2026 cluster of AI-agent-commerce standards, including ERC-8004 for agent identity and trust and ERC-8183 for agent-to-agent commerce primitives. LazAI and analysts at Bitget have grouped these into a two-layer model: 8004 governs how autonomous agents operate, while 8028 governs how the data and model assets those agents trade are owned and monetized.
​

ERC-8028 is a draft Ethereum token standard that wraps an AI dataset, model, or inference result inside a single onchain token carrying four fields: a class identifier, a usage value, a revenue-share ratio, and an optional expiry. Each token bundles ownership, metered access, and automatic payout routing in one structure, replacing the off-chain license agreements that typically govern AI assets.
​

The standard was proposed by LazAI, a project building decentralized AI infrastructure, and published for community review on Ethereum Magicians in late 2025. The corresponding code proposal lives in PR #1219 of the ethereum/ERCs repository. As of mid-2026 the EIP remains in draft and has not been finalized, audited as a standard, or deployed at production scale.
​

ERC-8028 extends ERC-3525, the semi-fungible token framework that introduced the ID, slot, and value triple scalar model. Where ERC-3525 is general purpose, ERC-8028 narrows the field schema to four AI-specific attributes and ties them to a revenue distribution function.
​

ERC-8028 was proposed by LazAI, the team behind a decentralized AI data and inference network. The author submitted the specification through Ethereum Magicians discussion and opened pull request #1219 against the ethereum/ERCs repository. As of June 2026, the EIP carries draft status, has not been merged as a final standard, and has no production-scale deployments outside LazAI's own testbed.
​

LazAI positions DAT as the native asset format for an onchain AI economy in which datasets and models become tradeable, revenue-generating instruments rather than static files behind API keys. The same team has published explanatory writeups framing 8028 as the asset-economics complement to 8004's agent-operations layer.
​

Independent coverage of the proposal has come from Bitget News, which described the standard as a candidate native AI asset format, and from Ethereum Magicians threads where commenters have probed gas costs, off-chain storage assumptions, and overlap with existing ERC-1155 designs. The draft has not been ratified, and material changes to the field schema remain possible before any final status.
​

ERC-8028, ERC-8004, and ERC-8183 form a three-layer stack for onchain AI commerce. ERC-8004 standardizes agent identity, reputation, and verifiable operations. ERC-8028 standardizes the data and model assets agents transact. ERC-8183 standardizes the commerce envelope that lets agents quote, pay, and settle with each other. Together they cover identity, asset, and transaction layers.
​

The three drafts emerged from different authors and serve different purposes, but they compose. An autonomous agent identified under ERC-8004 might purchase access to a dataset minted as an ERC-8028 token, then sell its inference output to another agent through an ERC-8183 commerce message. Each standard targets a different verb in the agent economy: operate, own, transact.
​

This layered framing matters because no single standard solves the full AI-on-blockchain problem. Builders evaluating the space need to track all three drafts and understand how a deployment that uses one often implies a need for the other two.
​

A Data Anchoring Token works by minting a single onchain token that holds four attributes: a class identifier marking what the asset is, a value field representing remaining usage quota, a share ratio defining how revenue routes back to contributors, and an optional expiry timestamp. When an application consumes the asset, the contract decrements value and distributes any attached payment by share ratio.
​

The lifecycle proceeds in three concrete steps. First, a dataset owner mints a DAT, setting class to identify the dataset, value to the number of inference calls authorized, share ratio to the contributor split, and expiry to a license deadline. Second, an AI service consumes one inference call, debiting value by one unit and emitting a usage event. Third, the payment attached to that call is auto-distributed across contributor wallets according to the recorded share ratio.
​

Large data payloads stay offchain. The standard anchors a cryptographic hash to the token so that any consumer can verify that the file delivered from storage matches the asset they paid for. This split between onchain commitment and offchain payload keeps gas costs tractable while preserving integrity guarantees.
​

Data provenance is the record of where a dataset came from, who touched it, and how it was transformed. For AI assets, provenance determines whether a model can be trusted, audited, or licensed for regulated use. Anchoring provenance onchain through ERC-8028 produces a tamper-resistant audit trail that persists across organizational boundaries and ownership transfers.
​

In current AI practice, provenance often lives in spreadsheets, contract PDFs, and platform logs, none of which a downstream consumer can independently verify. When a training dataset moves through cleaning, augmentation, and licensing steps, that chain typically becomes opaque. Regulators in the EU, under the AI Act, and in the US are pushing toward documented training-data lineage for high-risk AI systems.
​

ERC-8028 addresses this by writing every transfer, usage event, and revenue distribution to the chain as part of normal token operation. The provenance record is not an extra artifact someone has to maintain. It is a byproduct of using the token. That makes it materially harder to fabricate or lose, and easier for auditors and downstream model trainers to verify.
​

Cross-chain settlement matters for ERC-8028 because revenue-share payouts under the standard are typically many small payments to contributor wallets scattered across multiple chains. A dataset with hundreds of contributors will generate continuous micro-distributions every time an inference call is paid. Moving stablecoin value across chains efficiently is a different problem from defining the token itself, and the standard does not solve it.
​

This is where ERC-8028 makes AI datasets into revenue-sharing tokens, but every payout is a cross-chain stablecoin micro-settlement, and that rail is what Eco builds. Existing cross-chain options for stablecoin movement include Circle's CCTP, Hyperlane messaging, LayerZero, and Wormhole. Each handles cross-chain transport in a different way, and a DAT contract could integrate any of them at the payout layer.
​

The economic case for thinking about settlement up front is the stablecoin float that already exists. Per DeFiLlama data as of June 2026, the stablecoin market sits near $315B, with USDT at roughly $187B and USDC at roughly $76B. Revenue routed by ERC-8028 contracts will denominate in those assets, and contributors will want delivery on whichever chain they custody from.
​

ERC-8028 differs from earlier token standards because it is not general-purpose. ERC-20 is for fungible currencies, ERC-721 is for unique NFTs, ERC-1155 is for batched mixed inventories, and ERC-3525 is a general semi-fungible base. ERC-8028 narrows the ERC-3525 schema to four AI-specific fields and bolts in usage decrementing and automatic revenue split, behavior the earlier standards leave to application code.
​

ERC-20 cannot represent a dataset because every token in a pool is identical and indivisible at the attribute level. ERC-721 can represent a dataset as a single NFT but cannot fractionalize it among contributors without wrapping contracts. ERC-1155 enables batched mixed inventories but lacks usage metering. ERC-3525 supports the slot-and-value structure but has no opinion on usage quotas, revenue shares, or expiry.
​

The tradeoff is opinionation. ERC-8028 ships with assumptions about how AI assets should behave. Teams whose AI workflows do not fit the usage-and-revenue-share pattern may find a thinner standard like ERC-3525 or a custom ERC-1155 extension easier to work with.
​

Real-world use cases for ERC-8028 cluster around four patterns: shared training-data marketplaces, model access metering, AI-as-a-service quota delivery, and contributor revenue distribution for collaboratively built models. Each pattern depends on having ownership, usage, and payout encoded in the same token so that the asset itself enforces the business rules rather than a platform sitting in front of it.
​

Training-data marketplaces are the most-cited application. Multiple data providers mint a single DAT representing a curated dataset, with the share ratio recording each contributor's proportion. When a model team buys access, the value field decrements with each training run and revenue distributes automatically to provider wallets.
​

Model access metering uses the same mechanic for inference rather than training. An AI service provider mints a DAT that grants a buyer some number of inference calls. The token doubles as both license and quota counter. When the value hits zero, access ends without anyone needing to maintain a separate billing system.
​

Collaborative agent development represents a forward-looking case. An autonomous agent identified under ERC-8004 could hold a basket of ERC-8028 tokens representing the datasets, models, and tools it depends on, with revenue from its outputs flowing back through those tokens to upstream contributors. This is the agent-economy thesis that LazAI and the broader 8004 community have advanced.
​

Builders evaluating ERC-8028 should weigh three concrete factors: gas cost at expected usage volume, the maturity of supporting tooling, and the legal treatment of tokenized data licenses in their jurisdiction. The standard is still in draft, and production deployments outside LazAI's own ecosystem remain limited as of mid-2026.
​

Gas cost matters because every usage event is an onchain write. For high-volume inference services, recording every call on Ethereum mainnet quickly becomes uneconomic. Layer-2 networks such as Base, Arbitrum, and Optimism, or app-specific rollups, offer materially lower per-write cost. Batched accounting that settles to L1 periodically is another common pattern. The choice of execution venue should be made before the contract is finalized.
​

Tooling maturity affects how much engineering work falls on the implementer. Standard ERC-20 and ERC-721 deployments benefit from broad library support across OpenZeppelin, Hardhat, Foundry, and major wallets. ERC-8028 does not yet have equivalent library coverage. Teams adopting it will write more of their own scaffolding for minting interfaces, payout calculation, and indexer integration.
​

Legal treatment is the third variable. Tokenizing a license raises questions courts have not fully addressed: which jurisdiction governs the token, whether token holders have a contractual claim to off-chain data, and how disputes over share ratios resolve. Independent counsel and a clear off-chain license agreement that mirrors the on-chain token terms remain prudent.
​

Common questions about ERC-8028 cluster around four areas: how DAT differs from existing token types, whether it can be deployed today, whether it has been independently audited, and how revenue-share payouts actually reach contributor wallets. The short answers depend on the standard's draft status and the implementation choices each deployment makes.
​

How does DAT differ from an NFT? An NFT under ERC-721 is unique and indivisible. A DAT is semi-fungible, meaning tokens in the same class are interchangeable while different classes stay distinct, and the value field within a token can be partially consumed without destroying it.
​

How does DAT differ from an ERC-20 royalty token? ERC-20 royalty arrangements typically distribute pro rata to all holders of a single fungible token. DAT records share ratios at mint time and can route different fractions of payout to different addresses, with usage quota and expiry tracked alongside.
​

Can a team deploy ERC-8028 today? The standard remains a draft EIP. A team can implement the reference contract from PR #1219 on any EVM chain, but should expect interface changes before any final ratification and treat current deployments as production-experimental.
​

Has ERC-8028 been audited? The specification itself has been reviewed in public Ethereum Magicians discussion but has not been formally audited as a standard. Individual contract implementations require their own security review before holding any value.
​

Eco operates as cross-chain stablecoin settlement infrastructure, the layer that moves USDC, USDT, and other stablecoins between chains on user intent. The relevance to ERC-8028 is mechanical: when a DAT contract distributes revenue across many contributor wallets on different chains, the payout step is a cross-chain stablecoin movement, which is the problem class Eco Routes is designed for.
​

In a stack where ERC-8004 defines agents, ERC-8028 defines the assets they trade, and ERC-8183 defines the commerce envelope, settlement rails sit underneath all three. Multiple rails exist, including CCTP, Hyperlane, LayerZero, and Wormhole. Eco is one option among them, with its design centered on intent-based routing for stablecoin transfers across the chains where AI contributors and consumers actually hold balances.
​

For builders prototyping ERC-8028 deployments, the practical implication is that the token contract and the settlement contract are separate concerns. Choosing a token standard does not pick a settlement rail, and the two decisions should be evaluated independently against the chains, stablecoins, and contributor wallets involved.
​

Sources: ERC-8028 draft (Ethereum Magicians, PR #1219, ethereum/ERCs), ERC-3525 (EIPS), ERC-8004 (EIPS), Bitget News coverage of DAT, DeFiLlama stablecoin data as of June 2026.
​