# ERC-8454 Draft Adds a Bar Against Double-Counting Agent Claims

Published 2026-10-25 · Crypto.world Team · Crypto.world

> Draft ERC-8454 proposes registries for defining, measuring and valuing economic units that can include AI agents, tasks, models and financial positions. Its latest revisions block point-valued totals when a unit’s claims have not been shown to be distinct, while reviewers have raised an unresolved question about liability ranges.

A draft Ethereum standard for describing onchain AI businesses now bars a simple valuation total when the rights inside a unit may overlap. The latest revisions to ERC-8454, posted through October 24, add rules for deciding whether two positions represent the same claim, distinct claims or an unresolved pair. That distinction matters: adding the same revenue entitlement twice can turn a correct valuation into a misleading one.

Source [ethereum-magicians.org](https://ethereum-magicians.org/t/erc-8454-aifi-unit-framework/30040)

Gary Yang, author of the AIFi Unit Framework proposal, wrote: “The framework still does not decide whether two references are the same claim; it decides that addition needs someone to have said, with a basis, that they are not.” The draft remains under review, and a reviewer’s latest question points to a separate issue in how it bounds uncertain liabilities.

## ERC-8454 Defines the Economic Boundary

Yang posted the draft on October 21, and Ethereum assigned it ERC-8454 that day through pull request 2095. The proposal treats a “Unit” as an economic object defined by its boundary: it might contain an agent, skill, task, model, compute resource, data right, financial position or nested unit. Typed relations describe how those components connect, including ownership, licensing, obligations, revenue sharing and control.

The framework aims to make a valuation’s subject legible, rather than prescribe a single way to calculate its value. Each receipt records which version of a unit it measures, the assessment boundary, the economic interest, the snapshot, the method and a typed result. Results can include a monetary amount, an interval, a scenario, a score, a grade or an indication that evidence is insufficient.

Two registries carry the proposed records: AUFUnitRegistry stores units, versions, containment and acknowledgments; AUFAssertionRegistry handles characterizations, observations and valuations, including their validity, revocation and supersession. The draft depends on ERC-165, EIP-712 and ERC-1271, while leaving valuation functions and credit models to separate methods. It also specifies that missing observations are not zeros and internal transfers are not new flows.

## Overlapping Rights Block a Simple Sum

Reviewers pressed the proposal over two child units that might refer to one revenue entitlement through different contracts or descriptors. The first revision added “same-claim” and “distinct-claim” declarations, each supported by a stated basis. Subsequent changes clarified that missing or different references do not, by themselves, prove that positions are distinct.

Under the revised rule, every pair of financial positions within a unit must be grouped as the same claim, declared distinct with a basis, or treated as unresolved. A point-valued monetary total over an unresolved pair is non-conforming; a method may instead return an interval or insufficient evidence. The resolver must also report whether a declaration is merely claimed or has been endorsed by an account that controls the underlying interest.

The repository’s examples show the effect using a 40% revenue entitlement represented once as a token and once as a contractual right. With a supported same-claim declaration, the example counts the right once at 45.12. Without a declaration, it gives an interval from 45.12 to 90.24 or returns insufficient evidence, rather than selecting one figure as settled.

## A Reviewer Flags the Liability Case

The latest exchange raises a harder question about obligations. A reviewer asked how the interval rule handles two entries that might describe the same $100 debt: counting the liability once yields negative $100 for net equity, while counting it twice yields negative $200. Those outcomes reverse the intuitive ordering of “low” and “high” bounds.

The October 24 post asks whether a valuation method should calculate bounds across possible claim resolutions and return insufficient evidence when it cannot justify them. The proposal author had addressed several earlier review comments with revisions and tests, including rules tying an endorsement to the specific claim relation and unit version. The liability question is the latest unresolved point in the discussion.

## Testnet Evidence Predates the New Rules

The author reports a 40-transaction run on Ethereum’s Sepolia testnet on October 21, with both registries source-verified and every transaction successful. That run registered seven worked units and demonstrated signed valuation assertions, acknowledgments, supersession and revocation. It is evidence for an earlier snapshot, not a test of every later rule: the repository says the new claim declarations came after the run, and older unit records now expose unresolved-pair cases.

As of October 25, ERC-8454 is still a draft with editor review pending. The next test is whether the text can make its safeguards work for both assets and liabilities without presenting uncertainty as a precise figure; the review thread’s latest entry has not yet settled that question.

## Sources

1. Original source [Fellowship of Ethereum Magicians](https://ethereum-magicians.org/t/erc-8454-aifi-unit-framework/30040) ethereum-magicians.org

Other pages read

2. [GitHub](https://github.com/garyyang-finchip/aifi-unit-framework) github.com
3. [Ethereum ERCs](https://github.com/ethereum/ERCs/pull/2095) github.com

## Related stories

- [Phase launches mainnet with $45M backing for AI-agent connectivity](https://crypto.world/news/phase-mainnet-ai-agent-connectivity-45m)
- [Celestia’s Voyager Averages 21.2M TPS in Five-Hour Test](https://crypto.world/news/celestia-voyager-21m-tps-benchmark)
- [CFTC Chair Selig Puts Autonomous Trading on Regulatory Agenda](https://crypto.world/news/cftc-selig-autonomous-trading-regulatory-agenda)
- [Researchers Mutate Smart-Contract Tasks to Catch Copied Answers](https://crypto.world/news/researchers-mutate-smart-contract-tasks-catch-copied-answers)
- [Ethereum Draft Would Permanently Bar EIP-7702 Delegation](https://crypto.world/news/ethereum-draft-permanent-eip-7702-delegation-opt-out)

## Explore

[More articles](https://crypto.world/news)

---

Canonical: https://crypto.world/news/erc-8454-draft-agent-economic-units-double-counting
