A proposal from Infinihash KYT · Request for comment · v0.3 draft, October 2026
Crypto safety data that is free at the point of protection: the facts in a public log, the judgments left to a market.
Try the lookup in 20 secondsTry it · runs in your browser
This is the moment the whole proposal is for: a person about to pay a wallet that received stolen funds. Pick a destination, or paste any address, and press Send. Your browser hashes it, sends four bytes of the hash, and checks the answer itself.
waiting for Send–––A simulation: no network request is made, and the demo log holds one made-up claim, the illustrative record in Part 3. The real interface is drafted under The lookup.
Four seats are open. Each one makes the log real for everyone else. Pick yours and the email is already addressed.
Not any of these? The proposal needs argument more than agreement: select any sentence below and press Challenge this, or open the discussion.
Someone is about to send money to a wallet that received stolen funds last week. The fact that it did is known. It sits in a database that an exchange compliance team pays five or six figures a year to query. The person with their finger on Send does not have that database. Their wallet does not have it. The scam-report site they might check allows ten lookups a month on a free key. So the warning exists, and the one person who needed it never sees it.
That is the whole problem, and it is a market structure, not a technical gap. Three things hold it in place.
Labels are the product. Every forensics vendor sells the same thing: a list of addresses and what they are. Publish the list and the business is gone. No vendor will do it, and no one should expect them to.
A label is a liability. “This wallet is a scammer” is a statement about a person. Get it wrong and you are sued. The vendors carry that risk behind contracts and indemnities; a free list has nothing behind it, so a free list says nothing, or says it recklessly.
The free lists that were tried died. They had no revenue, no trust layer and no reason for anyone to keep writing to them. Bots and recovery scammers poisoned the open ones; the curated ones went quiet when the one volunteer moved on. Each failure is now cited as proof that open safety data cannot work.
Crypto is no longer only something people buy and sell on exchanges. Stablecoins settle at the scale of card networks: about $9 trillion moved in the year to late 2025 once bot activity is stripped out, more than half of Visa’s volume [1]. The fastest-growing use is not trading but paying. Payment firms handling real invoices report $122 billion a year in stablecoin payments, with business-to-business transfers the largest and fastest-growing part at $76 billion, up from $36 billion six months earlier [3]. In Nigeria, Turkey, Brazil, India and Indonesia people hold dollars in stablecoins and pay across borders with them [5]. Commodities are part of it: Reuters reports Russian oil sold to China and India settled partly in Tether, and Venezuela’s state oil company moved its sales to USDT under sanctions [9][10].
The consequence that matters here: more and more of that value never touches an exchange. Chainalysis records regular multi-million-dollar stablecoin transfers in energy and merchant trade between Africa, the Middle East and Asia, and notes that its own exchange data misses business-to-business and over-the-counter flows [7]. The BIS finds capital controls barely slow these flows [11]. Regulators now treat stablecoins as payment infrastructure, through the US GENIUS Act and Hong Kong’s licensing regime, and banks are running settlement pilots [13][14].
So the people transacting are increasingly transacting with each other, not with an exchange. Exchange-side compliance, which is where all the paid safety data lives, does not reach a direct transfer between two businesses or two people. Neither party gets a warning before it pays. The BIS itself warns that stablecoins are weak on identity checks [12]. The gap this proposal closes is not shrinking. It grows with the trade.
In October 2026 a $1M proof-of-funds theft was traced across four chains. It belonged to a ring of about 46 thefts, about $16M, running since 2022. The trace mapped 102 wallets and 64 exchange accounts. Every exchange the ring used had bought the paid data. Almost none of the ring’s wallets carried a public warning until the victim’s team filed them, one form at a time, while the ring kept moving money through fresh addresses that no form had reached yet.
Multiply that by every ring, and now by every business paying an invoice in USDT and every person sending dollars home. People say they do not trust crypto because things like this happen and nobody keeps the funds safe. The data that would keep them safer already exists. It is just priced for the wrong buyer.
Spam blocklists, breach-notification services, certificate transparency, CVE: each publishes the evidence for free and lets a market sell the judgment on top. Crypto does it backwards. It sells the evidence. This proposal turns it the right way round.
Public scam reporting exists. Chainabuse takes reports from anyone and is read widely. The Security Alliance’s sharing networks move warnings between security teams quickly. Exchange and vendor lists are deep and well funded. None of that is dismissed here; the log is designed to sit underneath all of it.
What none of them provides: a claim that is signed by an accountable issuer, tied to hashes of the evidence behind it, entered in a public record that no operator can quietly edit, with a dispute process whose outcome is as public as the claim, and an erasure path for the person who was wrongly named by circumstance. Reports can be anonymous and are hard to test. Membership lists are closed. Vendor lists are products. A fact that a victim’s counsel needs to rely on, in front of a court, has to be attributable, evidenced and correctable in public. That is the gap this proposal fills, and it is a narrow one on purpose.
Operators of existing services are invited to mirror the log and to file to it. If the record becomes substrate for the tools people already use, it has done its job.
Split the data in two. The facts go in a public, append-only log that anyone can read for free, forever. The judgments stay with whoever is making the decision: the wallet, the exchange, the vendor. The log never says “bad wallet”. It says “funds from this transaction are proceeds of this incident, here is the evidence, here is who says so and how sure they are.” Everything that follows is a consequence of that split.
Exchanges, forensics firms, police units and accredited victim-side investigators. Each holds scoped authority under a root body that sets the evidence standard and can suspend an issuer. Anyone can report; only an issuer can file. That is the trust layer the dead lists never had.
Transaction hash, address, amount, incident reference, and hashes of the evidence bundle (the victim’s statement, the police reference, the trace). Confidence is a number the issuer signs its name to. A claim with no evidence hashes is rejected at the door.
Small, flat, per claim. It funds the operators and it makes bulk poisoning expensive. Reading stays free. Victims’ counsel will pay it gladly: a logged claim is the thing they need next.
At least three independent operators (a university host, a security collective, an exchange) each sign the entry into their copy of the log. Public monitors compare the copies. An operator that quietly drops or alters an entry is caught by the other two. This is certificate transparency, applied to stolen funds.
A wallet about to sign sends the first few bytes of a hash of the destination address, gets back the matching bucket, and checks locally. The log never learns which address was queried. Free at consumer volume; a paid tier for heavy commercial use.
The log returns claims, not verdicts. The wallet decides what to show: a warning at confidence 6 from one issuer, a hard stop at 8 from two. The policy sits with the wallet, and so does liability for the warning it chooses to show. The split does not make the log liability-free, and it is not presented as a shield. An issuer is responsible for the truth of what it signs. An operator is responsible for publishing the record faithfully, including corrections and dispute outcomes. The evidence hashes, the dispute process and the erasure path exist because that responsibility is real, not as decoration.
When an issuer sends an on-chain notice to a wallet holding stolen funds, the notice transaction is logged. Anyone who later receives those funds can be shown to have had notice available. Victims’ counsel need exactly this to argue knowing receipt. Today it is reconstructed by hand, case by case.
An exchange gets the bulk feed by signing one agreement, in two steps. Step one is deliberately cheap: it files only the deposit addresses it has itself confirmed as receiving scam or theft proceeds, and in return receives the full feed, with a service level and no licence fee. Step two, once it has seen the feed's value: it adds the sweep mapping from each deposit address to its hot wallet. If it stops filing, the feed stops. The largest holders of safety data become its largest contributors, and the log fills the one gap every list has: which account the money finally reached.
Disputes. A fixed window, an evidence standard, a panel that excludes the issuer. The outcome is itself a log entry, so a correction is as public as the claim it corrects. Erasure. A person who proves the address is theirs and the funds are clean gets a superseding entry and, where the law requires it, removal of identifiers. Suspension. An issuer whose claims keep losing disputes loses its authority. The root body is the only party that can grant or remove it, and its decisions are logged too.
The window is thirty days from filing, shortened to seven where the claim is attached to funds still moving. A dispute pauses nothing by itself; it marks the claim as disputed in the record, and consumers set their own policy for disputed claims. The panel decides within thirty days of the window closing. Every step is an entry.
Filing fees will not fund this at the start, and the table below describes the steady state, not the launch. The bootstrap is a subsidy: public-goods grants and recovery pledges from firms that expect to use the record, covering operator costs and dispute panels for the first two years, with the amounts and the funders published. The log runs at a loss while it proves itself. That is stated here because a funding model that only works on paper is how the earlier lists died quietly.
| Layer | Free to | Paid by |
|---|---|---|
| The log, receipts, monitors | Everyone | Issuer filing fees, institutional reciprocity dues, recovery pledges, public-goods grants |
| Hash-prefix lookup API | Wallets and apps at consumer volume | A volume tier for heavy commercial use |
| Scores, clustering, exposure, SLAs | Nobody | Commercial vendors built on top of the log |
| Legal-grade evidence bundles | Victims, at cost | Counsel; recoverable as costs |
One more source, opt-in and in need of legal structuring: an issuer may attach a pledge that a small share of funds recovered on a logged claim goes to the operator foundation. Funding then scales with usefulness, not with how many exchanges renew a licence.
| What killed the open lists | What this has instead |
|---|---|
| No revenue path | Filing fees, reciprocity dues, recovery pledges, vendor service tiers. None of it from selling labels. |
| No trust layer; poisoned by bots and recovery scammers | Accredited issuers, an evidence standard, filing fees, disputes. |
| Nobody with a reason to keep writing | Victims’ counsel, who need logged notice to prove knowing receipt. |
| Nobody with a reason to keep reading | Exchanges, who want the map of which accounts received the money. |
| A single owner who could quietly remove entries | Append-only, several operators, public monitors. |
Nothing here is invented. Each piece has run for a decade or more somewhere else. The new part is combining them for stolen-funds provenance.
This is the concrete thing we propose to build, written as a first draft so it can be argued with line by line. Everything in this part is open to change in v0.4.
One schema, versioned, signed. Addresses and hashes below are illustrative.
// provenance claim, schema v0.1 (draft) { "claim_id": "pt:2026-10-22:000418", "kind": "proceeds_of_theft", // | proceeds_of_fraud | notice_delivered | dispute | withdrawal "incident": "inc-2026-1014-pof", "chain": "ethereum", "tx": "0x3c54…5cb0", "address": "0xa91c…6f43", "amount": {"asset":"USDT","value":"1000000.79"}, "evidence": ["sha256:9f2c…e41a", "sha256:71bd…0c88"], "issuer": "did:issuer:accredited-0007", "confidence": 8, // 1–10, signed by the issuer "status": "active", // active | disputed | withdrawn | superseded "supersedes": null, "notice_tx": "0xad19…b6f8", // on-chain notice delivered, if any "issued_at": "2026-10-22T17:46:53Z", "signature": "ed25519:…" }
What is deliberately absent: any name, any KYC identifier, any statement about who controls the address. The record is about a transaction and the money in it.
A confidence score is only useful if two issuers mean roughly the same thing by it. The rubric below is a first draft; the evidence standard, once set by the root body, owns it.
| Score | Meaning |
|---|---|
| 9–10 | Filed in or referenced by a court proceeding or a law-enforcement report, with the full evidence bundle hashed into the claim. |
| 7–8 | Investigator trace plus a victim statement, both hashed. Two independent sources agree on the flow of funds. |
| 5–6 | A single credible source: a trace or a victim statement with supporting records, hashed. |
| 3–4 | A report with partial corroboration. Published, but no consumer should hard-stop on it alone. |
| 1–2 | Uncorroborated. Reserved for notice-delivered records and early warnings clearly labelled as such. |
An issuer whose scores repeatedly fail to match outcomes in disputes is miscalibrated, and miscalibration is a ground for suspension, the same as filing without evidence.
At launch, Infinihash holds the root: it accredits the first issuers against the published evidence standard and can suspend them. That is centralisation, and pretending otherwise would cost more trust than admitting it.
The exit is part of the design, not a promise about intentions. Every accreditation and suspension decision is a public log entry from day one, so the record of how the root was used exists whether or not Infinihash keeps it. The root passes to a body with at least three members, no one of them Infinihash, when two conditions are met: three independent operators are live, and at least five accredited issuers are filing. Until then, Infinihash publishes each quarter how many operators and issuers are live, so progress toward the handoff can be checked by anyone. A single owner is a paywall waiting to happen; a temporary steward with published conditions is a bootstrap.
// draft API shape, hash-prefix query (k-anonymity: the log never sees the address) GET /v1/lookup/{chain}/{sha256_prefix_4_bytes} → { "bucket": [ { "address_hash": "sha256:…", "claims": [ claim_id, … ], "max_confidence": 8 }, … ] } GET /v1/claim/{claim_id} // full record, free GET /v1/incident/{incident_id} // all claims under one incident POST /v1/claim // issuers only, signed, filing fee POST /v1/dispute/{claim_id} // anyone, with evidence; opens the window
The reference wallet integration shows one line when a destination matches: “This address received funds that an accredited issuer has logged as proceeds of theft (confidence 8 of 10, incident inc-2026-1014). Continue?” The wallet chooses the threshold and the wording. The log supplies only the fact.
An exchange that wants the bulk feed signs one agreement, in two steps. Step one: it files, as an accredited issuer, every deposit address it has itself confirmed as receiving scam or theft proceeds, within a fixed number of days of confirmation. In return it receives the full log as a feed, with a service level, and pays no licence fee. Step two, once it has seen the feed’s value: it adds the sweep mapping from each deposit address to its hot wallet, the record no single list has today. If it stops filing, the feed stops. The first step is deliberately cheap to accept; dependence on the feed is what makes the second step reasonable.
Infinihash builds the reference implementation and acts as first issuer. It will not own the log. A single owner is a paywall waiting to happen.
One interest, stated plainly: Infinihash intends to sell judgements built on top of this log. That is the business the log makes possible, for Infinihash and for every other vendor. The log is structured so that it never needs Infinihash to survive: independent operators, public monitors, and a root that leaves on published conditions. Judge the structure, not the motive.
Comments are open until 30 November 2026, in the GitHub discussion or by email to hello@infinihash.com. Say whether we may quote you. Each revision will say what changed, why, and who argued for it.
Disagree with a sentence? Select it anywhere on this page and press Challenge this: a GitHub thread opens with the sentence quoted and a link back to where it sits. Every section heading has the same link.
hello@infinihash.com