The risk layer for on-chain finance

Public blockchains settle value faster and more cheaply than correspondent banking, and they do it without the risk apparatus that made correspondent banking usable by regulated institutions. FIN-TECHAI builds that missing apparatus.

Four production modules read public chain data and return risk signals an institution can act on: Sentinel for screening and transaction monitoring, CreditGraph for repayment signals where no credit file exists, PayRoute for settlement routing that prices risk alongside cost, and Ledger Mind for reconciliation and evidence. All four are live today across seven networks.

Three commitments distinguish the design. Every output carries a confidence band, so an integrator can tell a strong finding from a weak one. Every output carries its contributing factors, so a decision made with our signal can be explained afterwards by the customer, to their regulator, without us in the room. And no output is ever presented as a verdict about a person — a score is an input to a policy the customer owns.

$FTAI is the settlement asset for that access. Usage of the risk layer is priced in $FTAI, holders govern the parameters described in section 14, and the token's supply is fixed at 1,000,000,000 units by the token contract. Sections 10 to 13 set out utility, supply, release schedule and the presale.

On-chain settlement removed the intermediaries. It also removed everything those intermediaries were doing that nobody had priced.

Settlement improved; diligence did not

Blockchain settlement collapses a multi-day correspondent chain into seconds, and all-in costs on on-chain rails commonly land between 0.1% and 0.5% against the 2–7% a traditional wire genuinely costs once fees, FX spread and intermediary deductions are counted. That gap is why institutional volume moved.

What did not move with it is the layer of judgement a correspondent bank supplied: knowing who the counterparty is, whether the funds are clean, whether the receiving institution has controls, and whether a payment should be made at all. On a public chain that judgement is technically possible — the entire history is public — and practically absent, because the data is unlabelled, cross-chain, and adversarial.

Three concrete failures

Screening happens too late. Many programmes screen after settlement, which produces an incident report rather than a control. A sanctions breach is strict liability in most jurisdictions; discovering it afterwards does not cure it.

Monitoring rules are inherited, not designed. Rules ported from wire-fraud systems fire constantly on benign on-chain behaviour — market-maker rebalancing looks like structuring, a bridge contract looks like a high-velocity counterparty — while missing the typologies that matter. Legacy rule sets routinely run above 90% false positives, and a queue nobody can clear is a control that is not operating.

Credit has no file to pull. Lending against on-chain collateral means underwriting counterparties with no bureau record, no income statement, and the permanent ability to start again at a fresh address.

Why a shared layer

Each of these is solvable in-house, and each is solved badly in-house, because the expensive part is not the model — it is the ingestion, normalisation and attribution work underneath it, replicated per firm at enormous duplicated cost. A shared risk layer amortises that work and, more importantly, improves with aggregate coverage: a typology observed at one customer strengthens detection for all of them, without any customer's data being disclosed to another.

Four constraints shaped every part of the system, and each rules out a design that would otherwise have been easier to build.

1. A signal, never a verdict

Outputs describe addresses and transactions, not people. No output is a determination of wrongdoing, and neither we nor our customers may present one as such. This is partly an ethical position and partly a legal one: where an automated process produces a consequential outcome for an individual, that individual is frequently entitled to an explanation and to human review.

2. Confidence is part of the output

A score with no stated confidence is an assertion. Thin history, young coverage on a network, and low-certainty attribution all widen the band rather than being hidden inside a single number. A score of 72 at high confidence and a score of 72 at low confidence are different findings and must not trigger the same action.

3. Explainability is a deliverable, not a feature

Every response carries its contributing factors, the attribution source behind each, and the list versions queried. A customer must be able to reconstruct why a payment cleared or an alert fired two years later, from their own records, without us.

4. Exposure is a gradient

Direct receipt from a sanctioned address is a fact about a counterparty; a five-hop path to the same address is a fact about network connectivity. Every exposure finding carries hop distance and share of value, because collapsing them into a boolean is the single largest manufacturer of false positives in this category.

The system is deliberately boring in shape: ingest, normalise, build a graph, score the graph, serve the score. The engineering value is concentrated in the second and third stages, which is where most implementations are weakest.

Separating ingestion from scoring

Coverage started on EVM chains, which share an execution model: one ingestion path, one log format, one way to trace a token transfer. Solana broke that assumption, and the fix was to split ingestion from scoring entirely. The benefit is that adding a network widens the graph the models read without changing what a score means — which matters to any customer with thresholds tuned against it.

Normalisation is where correctness is won or lost

Two examples of why this stage carries the risk. Tron permits resource delegation, so one account can pay another's costs; a tracer that reads "paid the fees" as evidence of common control will merge a service and all of its users into one cluster, and every signal computed over that cluster is meaningless. Separately, a Tron address has a hex form whose tail is indistinguishable from an EVM address, so a naive identifier space serves an Ethereum verdict for a Tron transfer. Every identifier in our graph is namespaced by chain, and delegation is a distinct edge type weighted separately from value flow.

Serving and evidence

The API layer is designed around one requirement that shapes everything else: the response is the customer's evidence. Every response therefore carries a trace identifier, the list versions queried, and per-finding attribution and confidence, and is intended to be persisted whole rather than reduced to a decision. A customer who stores only "score 34, threshold 40, allowed" has a policy statement; a customer who stores the response has a record.

All four are in production today. Each answers a different question against the same graph.

6.1 Sentinel — screening and monitoring

Sentinel screens addresses and transactions before value moves, and monitors flow continuously afterwards. It resolves sanctions and watchlist matches, attributes counterparties to services, and measures exposure to mixers, darknet clusters and known ransomware addresses with hop distance and value share attached to every finding.

Detection runs on typologies specific to virtual assets: layering across fresh addresses, chain-hopping through bridges and swaps, mixer exposure, structuring below reporting thresholds, and interaction with services that have weak or absent controls. Re-screening runs on a risk-banded schedule and on trigger events — a list update, a new attribution on any cluster in a customer's graph, adverse media, or a material shift against the customer's own behavioural baseline.

6.2 CreditGraph — repayment signals

CreditGraph produces a repayment signal for addresses with no off-chain credit file. The strongest single input is behaviour under stress: what an address did during a sharp drawdown with liquidation close, which is recorded permanently and timestamped for anyone to read. Collateral provenance is the second — whether assets were held for a year or acquired through a loop or a related-address transfer minutes before credit was drawn, a distinction invisible in a balance snapshot.

A thin address is scored unknown with a wide band, never bad. Conflating absence of evidence with evidence of poor credit is the most common error in on-chain underwriting, and the model is built to refuse it.

6.3 PayRoute — settlement routing

PayRoute selects a settlement corridor — the full path from the payer's asset and chain to the payee's, including the converting venue, the rail that moves value, and the endpoint that settles into local currency. It scores every hop and chooses against a policy the customer sets, optimising for all-in cost including slippage, time to final settlement rather than first confirmation, and counterparty risk at each hop.

When no acceptable corridor is available, PayRoute holds the payment and reports why, surfacing each rejected alternative with its reason. It never silently falls back to a route the policy would have rejected. Every decision retains the full candidate set, which is the field that tells an auditor whether the policy was working rather than merely what happened.

6.4 Ledger Mind — reconciliation and evidence

Ledger Mind reconciles on-chain movement against a customer's internal ledger and assembles the evidentiary record the other three modules generate: alert disposition and rationale, reviewer identity, policy version in force, and the full response bodies behind each decision, held immutably for the applicable record-keeping period.

Inputs

Every score derives from public chain history: address age, the shape of the funding graph, hop distance to sanctioned or exploited clusters, the mix of contracts and services interacted with, and behavioural similarity to addresses later found to be involved in theft or laundering. No input is decisive alone. A new address is not suspicious for being new; every address was new once, and legitimate users create fresh addresses constantly for sound privacy reasons.

Output shape

Reading the bands

Why scores move

A score is a property of an address at a point in time, not of the address. It changes when new flow arrives, when an attribution is published on a cluster the address touches, when a counterparty is designated, or when a model is retrained — any of which can move a score while the address itself does nothing. The operational consequence for integrators is that a cached score is not a screening result: enrichment may be cached for hours, a score for minutes, and a sanctions determination for seconds or not at all.

Seven networks are in production. Coverage is not the same as parity, and the output says which is which.

Heuristics tuned on years of EVM history are younger on Solana and Tron, and the confidence band widens accordingly rather than implying a precision that has not been earned. The question worth asking of any analytics provider, including this one, is not whether a chain is covered but what the confidence distribution on that chain looks like against Ethereum, and how long it has been in production.

Cross-chain tracing resolves automatically where a flow crosses networks. This matters because illicit actors rarely stay on one chain — value moves through bridges, swaps and privacy tooling in minutes — and a stack that leaves an analyst manually stitching hops together has coverage on paper only.

Treasury control

Treasury funds sit behind a multi-signature wallet. No single key can move value and no single person holds enough keys to reach the threshold alone; signers are separated by role and by device, so compromising one workstation does not compromise a quorum. Signer and device diversity is treated as seriously as signer count — five signers on identical hardware, firmware and machine images is a single point of failure wearing a quorum's clothing.

Movements fall into three classes, each with a different path:

The first-time-destination trigger does disproportionate work: most treasury losses are not exotic cryptographic failures but payments to an address that looked right. A timelock does not prevent a malicious transaction — it makes one visible while there is still time to react, which is only a control if queued transactions are alerted on, reviewed by a named owner, and cancellable through a rehearsed procedure.

Platform security

Encryption in transit and at rest, least-privilege access control, multi-factor authentication on administrative systems, network segregation, logging and monitoring, and periodic penetration testing and review. Where a breach is likely to risk individuals' rights, the applicable supervisory authority is notified within 72 hours of becoming aware and affected users without undue delay.

Audit and published artefacts

$FTAI is the settlement asset for access to the risk layer. Its demand is a function of usage, not of narrative.

Utility

Metered access

API consumption across all four modules is priced and settled in $FTAI. Screening calls, scoring calls, routing decisions and reconciliation jobs each carry a unit price set by the parameters in section 14.

Access tiers

Held balance determines rate limits, re-screening frequency, cross-chain trace depth and evidence retention beyond the statutory minimum.

Governance

Holders vote on the parameter set described in section 14, weighted by held balance and subject to the participation floor set there.

Model contribution

Typology submissions and attribution corrections that survive review are rewarded from the community allocation, which is how coverage improves faster than a single team could drive it.

Token facts

Total supply is fixed by the token contract and is the one commercial figure in this document that is settled. Everything in sections 11 to 13 is indicative and subject to the final approved sale sheet.

Release schedules exist to make the float predictable. All schedules are enforced by the vesting contract, not by policy.

Claim mechanics

Contributing gives an entitlement to $FTAI, not immediate delivery. Tokens become claimable at token launch through the published claim interface, and are claimed from the same wallet that made the contribution — an entitlement cannot be reassigned to a different wallet. Unclaimed entitlements remain claimable for twelve months from token launch, after which they may return to the treasury.

Transfer restrictions

Until an entitlement has vested it may not be sold, pledged, assigned or otherwise transferred. Where tokens are sold to US persons under a securities-law exemption they are restricted securities for US law purposes, may not be resold except pursuant to registration or an available exemption, and are subject to the applicable holding period; the relevant allocation is legended or locked accordingly.

The presale allocation is sold in stages at published prices with published caps. Prices increase between stages; unsold allocation rolls forward.

Terms that apply to every contribution

• Minimum contribution — the equivalent of US$100 in a supported asset, measured at the rate applied at settlement.
• Price is set at settlement, not at submission. Where a contribution arrives as a stage closes, the price applied is that of the stage in effect when it settles on-chain.
• Quoted amounts are estimates based on third-party market rates, not guarantees of tokens received.
• Supported assets only, on their native chain, to a sale address published through an official channel. Assets sent in another token, on another chain, or to another address are unrecoverable.
• Contributions are final. There is no cooling-off period and no refund, save where the presale is cancelled entirely or a refund is legally required. Mandatory consumer rights that cannot be excluded are unaffected.

Eligibility

Contributors must be at least 18 and act on their own behalf. The presale is closed to persons in Cuba, Iran, North Korea, Syria, Russia, Belarus, Myanmar, Afghanistan, Ukraine and Venezuela. US persons may participate only under the exemption stated in the terms of service and only where they meet the applicable investor standard, with participation additionally restricted in some states. EU and UK residents may participate; nothing in this document is a prospectus or an approved financial promotion. Full eligibility conditions are in clause 03 of the terms of service.

Governance covers the parameters that determine what access costs and how the layer behaves. It does not extend to matters the Foundation is legally obliged to decide itself.

In scope

• Unit pricing for metered API access across the four modules.
• Access-tier thresholds: rate limits, re-screening frequency, trace depth, retention beyond statutory minimums.
• Emission rate and reward weighting for the community allocation.
• Priority ordering for new chain coverage.
• Ecosystem grant allocation from the ecosystem tranche.

Out of scope

Sanctions screening thresholds, list sources, model risk decisions, AML policy, treasury signer composition and any matter subject to regulatory obligation are outside governance. These cannot be delegated to a token vote: they carry legal accountability that sits with the Foundation and with each customer, and a vote cannot absorb strict liability.

Mechanics

Proposals require a sponsor holding at least 0.5% of circulating supply, a seven-day discussion period before a vote opens, and a five-day voting period. A vote carries on a simple majority of votes cast subject to a 4% participation floor of circulating supply. Passed proposals affecting on-chain parameters execute behind the structural timelock described in section 9, so every change is visible before it takes effect.

Benefits attaching to presale wallets — entry in the founding registry, governance participation and any presale-holder allocations — attach to the wallet that claims, not to the tokens. Selling, bridging or transferring a position does not transfer them.

Statements below describe present intent. They are not commitments, and timing may move for technical, security or regulatory reasons.

This section summarises the principal risks. It does not replace clause 11 of the terms of service, which contains the full risk disclosure and forms part of the agreement governing any contribution.

Token and market

• The price of $FTAI is volatile and may fall to zero. You may lose the entire value of your contribution.
• A liquid market may never develop. Listings are not guaranteed, and vesting means a holding may be illiquid for a prolonged period.
• No government compensation or deposit-guarantee scheme covers a contribution.

Legal and regulatory

• The legal and tax character of $FTAI is unsettled. A regulator or court — including in the United States, the European Union or the United Kingdom — may treat it differently from the description in section 1, which may restrict your ability to hold, claim, transfer or realise value from it.
• Regulatory change may restrict or prohibit the token or the services in your jurisdiction.
• Where tokens are acquired under an exemption, resale restrictions apply.

Technical

• Smart contracts may contain defects notwithstanding audit. An audit is not a warranty of security.
• Blockchain networks may congest, fork, halt or fail, and transactions cannot be reversed.
• Losses from phishing, impersonation, malicious approvals, compromised devices or funds sent to an unverified address are the contributor's own. We never contact contributors first to request a seed phrase, private key, signature or payment.

Model and product

• A score or signal may be wrong in either direction. Attribution is revised as new intelligence arrives, and no score is a determination of wrongdoing.
• Coverage maturity varies by network; confidence bands reflect this but do not eliminate it.
• The roadmap is a statement of intent. Features, timing and third-party integrations may change or not be delivered.

Execution

• The token launch date is a target and may move. Stage timing may change, stages may close early, and the presale may be cancelled for legal, security or operational reasons.
• The Foundation depends on third parties it does not control, including networks, wallet providers, price feeds, custodians, auditors and infrastructure hosts.

Attribution

The association of an address or cluster with a real-world service or entity, carrying a source and a confidence level.

Chain-hopping

Moving value across two or more networks via bridges or swaps, typically to break a trace.

Cluster

A set of addresses inferred to be under common control. Inference requires corroboration; fee payment alone is not evidence of control.

Confidence band

The stated certainty of an estimate. Widens with thin history, young network coverage or low-certainty attribution.

Corridor

The complete path a payment takes from the payer's asset and chain to the payee's, including converting venue, rail and settlement endpoint.

Exposure

A measured relationship between an address and a category of risk, always qualified by hop distance and share of value.

Hop distance

The number of transfers separating two addresses. Direct exposure and five-hop exposure are materially different findings.

Settlement finality

The point at which a transfer is irreversible. Distinct from first confirmation, and the figure that matters for routing.

Typology

A recognised pattern of illicit behaviour encoded as detection logic — layering, structuring, mixer use, chain-hopping.

Unhosted wallet

A wallet no regulated provider custodies. Transfers to one carry a duty to verify the customer controls the address.

VASP

Virtual asset service provider — an entity conducting virtual-asset business subject to AML obligations.

$FTAI

The utility token used to pay for access to the FIN-TECHAI risk layer and to exercise the governance rights in section 14.