Verifiable Contribution Without Global Deanonymization

#privacy #opensource #web3 #ai

AI changes the cost structure of public work. Code, summaries, tests, listings, translations, and operational decisions are becoming easier to produce. That is good. It also means that the scarce signal may move from "who can write the artifact" to "who actually participated, what did they do, what evidence exists, and what responsibility did they accept?"

This is the design question behind WebAZ: how can a protocol make real participation verifiable for humans and future AI agents without turning contribution records into a global identity trap?

The naive answer is to bind every action to a single strong identity. That is tempting because it makes abuse and attribution easier to reason about. It is also dangerous. Vitalik's recent writing on ZK-wrapped identity makes the core problem clear: even when zero-knowledge proofs hide legal details, a one-person-one-identity system can still reduce practical pseudonymity, create coercion risks, and push people toward a single public life. AI makes the problem sharper because writing style, timing, topic selection, social graph, and workflow traces all become easier to correlate.

Contribution systems face the same trap. If the future of work is agent-assisted, then many valuable actions will happen through mediated workflows: a human asks an agent to inspect an issue, draft a patch, compare an order state, prepare a listing, or check a policy. Some of those actions should become durable records. But if every record becomes globally linkable by default, the contribution layer starts to look less like public memory and more like surveillance.

The alternative is not to abandon verifiability. The alternative is to attest actions, not souls.

A contribution attestation should answer a narrow question: "Did this action happen, under what rules, with what evidence, and who accepted accountability for it?" It does not need to answer every identity question at once. In many cases the useful record is not "this legal person did everything under one permanent identifier", but rather:

this task was proposed;
this agent or tool produced an artifact;
this human account authorized or accepted it;
this review outcome happened;
this payment or order state changed;
these pieces of evidence support the claim;
this disclosure is sufficient for this context.

That framing lets different layers have different visibility. Public records can preserve hashes, timestamps, task references, status transitions, and review outcomes. Private or selectively disclosed records can preserve richer evidence, identity links, or payment details. A user should be able to prove enough for a specific interaction without carrying their entire public history into every room.

This matters for AI agents because agent work needs attribution without pretending the agent is the person. If a person delegates work to an agent, and the work creates value, the system should be able to record that delegation and outcome. But the accountable party should be explicit. A mutable admin seat, a GitHub handle, a wallet, a passkey credential, and a human contributor are related, but they should not be collapsed into one universal identity object. Collapsing them makes the system simpler at first and more brittle later.

It also matters for commerce. WebAZ is exploring programmable USDC payment flows and non-custodial direct settlement alongside contribution records. Those two kinds of state should not be confused. Payment state should be compact and bounded: who is paying whom, what confirmation occurred, what risk acknowledgements were made, and what changed in the order. Contribution or reputation state should be evidence-oriented and contextual: who helped build, verify, operate, or improve the system. A payment should not automatically become a permanent reputation claim; a contribution should not automatically reveal a payment graph.

A useful protocol for AI-era participation may therefore need five properties:

Contextual attestations. Records should say what they prove and what they do not prove.
Selective disclosure. Contributors should be able to reveal the part of their history that is relevant to a task, role, or dispute without exposing unrelated activity.
Pluralistic identity. The system should accept multiple identity anchors and evidence sources rather than enforcing a single universal credential.
Human-accountable delegation. Agent-assisted work should be recordable, but the human or organization accepting responsibility should remain legible.
Bounded economic state. Payments, order confirmations, contribution evidence, and identity claims should be connected only where the workflow requires it.

None of this is a finished answer. WebAZ is an open-source experiment, not a standard. The current hypothesis is modest: if AI makes artifact generation cheap, then protocols should make participation, responsibility, and evidence more legible. But that legibility must not require default global deanonymization.

The open questions are the interesting part:

What should be public by default, and what should remain private unless disclosed?
How should agent-assisted contributions be attributed without turning every tool call into identity exhaust?
Can contribution attestations be portable across communities while staying contextual enough to resist coercion?
Should some attestations decay, expire, or become less visible over time?
What is the smallest onchain footprint that still gives participants durable leverage?

The wrong outcome is a world where AI writes everything and humans disappear from the record. An equally wrong outcome is a world where every contribution becomes a permanent global dossier. The useful path is narrower: verifiable participation, selective disclosure, and accountability scoped to the action that actually matters.

References: