For fraud, payments & risk leadership
The login was real. The payment details were not.
Your controls asked whether the person was allowed in, and whether the activity looked unusual. Neither question catches a payment whose recipient details changed after it was authorized. CapchaCloud answers one question the rest of the stack does not: do the final payment details still match what was actually authorized?
CapchaCloud does not move funds, does not block payments, and does not decide anything. It returns a comparison and a receipt. Your people make the call.
Where the gap actually is
Business email compromise does not look like an intrusion, because it usually is not one. Nothing gets hacked. A legitimate person, correctly authenticated, is persuaded to send money to the wrong place — or a correct instruction is quietly altered somewhere between approval and execution.
- A real customer or employee logs in legitimately. Authentication passed. Credentials are genuine, the device is known, the session is clean. Nothing looks wrong, because nothing about the access is wrong.
- Between the authorization of a payment and its release, a detail changes. A recipient name, an account number, or an amount. It may change in an email thread, in a re-keyed instruction, in an attached invoice, or in a follow-up "correction" that arrives before release.
- Existing controls answer different questions. Authentication asks whether this person is allowed in. Risk scoring asks whether this looks suspicious. Case management asks how an alert should be investigated. All three are doing their jobs correctly.
- None of them answer the one that matters here: do the final payment details still match what was actually authorized? That is a comparison question, not an identity question and not a probability question.
- By the time the answer is assembled, the funds are often gone. The request, the approval, the recipient record, and the release record usually live in four different systems. Manually reconciling them is an after-the-fact investigation, and recall windows are short.
Four controls, four different questions
These are complementary controls, not competing ones. Authentication, risk analytics, and case management are each necessary, each mature, and each good at what they were built for. The table below is about scope, not quality — CapchaCloud is the fourth column, not a replacement for the first three.
Comparison of authentication, risk scoring, case management, and CapchaCloud by the question each control answers, its output, and its timing
| Control |
Question it answers |
Output |
Timing |
| Authentication & access |
Is this person who they claim to be, and are they permitted to act? |
An allow or deny decision on access, plus a session. |
At sign-in and at step-up, before the payment exists. |
| Risk scoring & behavioral analytics |
Does this activity look anomalous relative to expected patterns? |
A score, a rule hit, or a ranked alert for human review. |
Continuously, and at submission of the transaction. |
| Case management |
How should a raised alert be triaged, investigated and dispositioned? |
A workflow, an assigned owner, an investigation trail, a disposition. |
After an alert exists, on the investigator's clock. |
| CapchaCloud |
Do the final payment details still match what was actually authorized? |
A field-level match or mismatch, plus a sealed Evidence Record you can verify without trusting us. |
Between authorization and release, while the payment is still recallable. |
A note on honesty: nothing here computes a fraud score. CapchaCloud has no model, no classifier and no anomaly engine in this path. The comparison is deterministic — details are normalized and compared, and the result is a match, a mismatch, or a pending confirmation.
You do not have to trust us — that is the design
Most vendors ask a bank to accept their word about what happened. CapchaCloud is built so you never have to. Every Evidence Record is cryptographically sealed, and its integrity can be checked by your own people, on your own hardware, with no call back to CapchaCloud.
What is in the seal
- Content hashing. The record payload is hashed, so any later alteration is detectable rather than silent.
- Append-only ledger with Merkle inclusion. Records are folded into a hash-linked, Merkle-batched chain, so a record cannot be quietly inserted, removed, or reordered after the fact.
- Classical signatures, with post-quantum migration underway. Blocks are signed with Ed25519. The ledger additionally supports an ML-DSA-65 post-quantum co-signature alongside the classical signature where enabled for a deployment; the public explorer reports the live co-signer count, and the offline verifiers check a second signature whenever a block actually carries one. Algorithm-by-algorithm detail is published in the crypto inventory.
- Independent RFC 3161 timestamps. Each sealed block hash is submitted to multiple independent timestamp authorities — DigiCert, Sectigo and FreeTSA — and every returned token is retained as its own witness.
- External anchoring. The same block hash is submitted to multiple OpenTimestamps calendars and matures asynchronously to a Bitcoin attestation. Maturation takes time; it is not instant finality, and we say so.
How your team checks it, offline
The verifier is a standalone script with zero dependencies beyond Node 18 built-ins. Download it once, review the source, and run it against any evidence bundle. It re-derives the hashes and checks the signatures itself.
curl -O https://capchacloud.com/verify-evidence-bundle.mjs
curl -o bundle.json "https://capchacloud.com/api/v1/public/evidence-bundle/<id>"
node verify-evidence-bundle.mjs bundle.json
A second verifier re-derives every block in the append-only ledger independently. Walkthrough and live chain state: public chain explorer. Live anchor status: /api/v1/transparency.
This matters in a dispute. An Evidence Record that only CapchaCloud can vouch for is worth very little to a bank's counsel. One that your own examiner can re-derive from the bytes is a different kind of artifact.
The bank keeps every decision. Every single one.
This is the part procurement cares about most, so it is worth being blunt about it. CapchaCloud is not in your payment path, not in your ledger, and not in your authority chain.
It never moves funds
CapchaCloud does not execute, custody, hold, transmit, or settle money — for anyone, in any product. Every response from the verification gate carries a standing advisory note stating exactly that, so the boundary travels with the data instead of living only in a slide.
It never blocks a payment
A mismatch marks CapchaCloud's own record as not released and seals a receipt saying why. Nothing reaches into a payment rail, a core system, or a queue. Your systems remain the only thing that can stop or send a payment.
It never makes the call
CapchaCloud returns a comparison and a receipt. An authorized bank employee reads it and decides to release, hold, or escalate. There is no automated action, no auto-decline, and no place where our software substitutes for your judgment.
Shadow mode. You do not have to put this in the live path to find out whether it is worth anything. Run the comparison alongside your existing process, gating nothing and changing nothing: the same instructions flow through your current workflow untouched, while CapchaCloud independently records what it would have surfaced. You compare the two at your own pace, and decide on evidence rather than on a pitch.
What the output actually looks like
Illustrative — synthetic data
The record below is a rendering with invented values, shown to make the shape concrete. It is not a real customer record, not a real payment, and not a screenshot of a bank deployment.
Payment-detail comparison — authorization versus confirmation
Four states are sealed independently and cross-linked: submitted, verdict, confirmation, final.
Illustrative field-level comparison between details sealed at authorization and details independently re-stated at confirmation
| Field |
Sealed at authorization |
Independently re-stated at confirmation |
Result |
| Recipient name |
NORTHRIDGE FABRICATION LLC |
NORTHRIDGE FABRICATION LLC |
match |
| Recipient account |
••••4471 (SHA-256 of full number held; number itself never stored) |
••••9038 |
mismatch |
| Amount |
USD 184,200.00 (sealed at authorization) |
not re-stated by confirmer |
unchanged against sealed record |
| Confirming party |
— |
approver_ref: ops-4412 |
recorded |
The account comparison is a hash comparison. The confirming party types the account number independently rather than reading it back from the instruction, it is normalized and hashed, and the hashes are compared. The full number is never written down on either side of the comparison — only a SHA-256 hash and the last four digits are persisted.
Result: mismatch on recipient account. CapchaCloud's record is marked not released and a receipt is sealed stating the reason. No payment was stopped by CapchaCloud, and no funds were touched by CapchaCloud. The receipt goes to an authorized bank employee, who decides what happens next.
The reported scale, from primary sources
Only figures we can point at directly. Each is linked to the publishing body so your team can check it rather than take it from a vendor page.
FBI Internet Crime Complaint Center, 2025 Annual Report
- $20.877BTotal reported losses from cyber-enabled crime in 2025, up 26% year over year.
- $3,046,598,558Verified business email compromise losses, from 24,768 complaints. BEC is the second-largest crime type by total loss.
- 22,364AI-related complaints reported in 2025, with roughly $893 million in associated losses.
Source: Federal Bureau of Investigation, Internet Crime Complaint Center — 2025 Internet Crime Report (PDF). Figures are reported losses from complaints received by IC3; they are not an estimate of total losses economy-wide.
Deloitte Center for Financial Services projection
- $40B by 2027Projected US generative-AI-enabled fraud losses, up from $12.3 billion in 2023.
- 32% CAGRCompound annual growth rate implied by that projection over the period.
- ~$22BDeloitte's conservative scenario for the same 2027 horizon — we quote it because a range is more honest than a headline.
Source: Deloitte Center for Financial Services — Deepfake banking fraud risk on the rise. These are projections by Deloitte, not measurements, and not a forecast of losses this product would prevent.
Start with a retrospective audit, not a purchase
We would rather be measured than believed. So the first step is not a contract, a pilot, or an integration — it is an evidence exercise run entirely on history that has already happened.
What we ask for
A closed, historical window of already-resolved payment activity — authorization records and release records for a period that is finished and settled. Scope, field set, format, and any de-identification are yours to define before anything is shared.
What you get back
A count of authorized-versus-released detail mismatches we would have surfaced before release, in your numbers rather than our projections — with the specific records named so your team can check each one against what actually happened.
What it does not involve
- No production risk. The window is closed and historical. Nothing is running against live traffic.
- No live workflow changes. No integration into your payment path, no code in your release process, no configuration on your rails.
- No funds movement. There is nothing to move. Every payment in the window is already resolved.
- No commitment. If the count is zero, that is a real and useful answer, and we will report it as such.
On sizing arithmetic. If a retrospective audit surfaces mismatches, multiplying that count by an average payment value is illustrative arithmetic — it describes the value of the payments involved, and nothing more. It is not a forecast of losses avoided, not a claim about what this product would have prevented, and not a savings estimate. Whether any individual mismatch was fraud, a keying error, or a legitimate late correction is a determination for your investigators, not for us.
What this does not prove
CapchaCloud receipts carry their own limits with them — a standing advisory note on every verification-gate response, and an explicit does_not_prove list on the receipt types that publish one. A record that overstates itself is worse than no record, so the same discipline applies to this page.
- It does not move or hold funds. CapchaCloud never executes, custodies, holds, transmits, or settles money.
- It does not block payments. Nothing here reaches into a payment rail or a core banking system. Only your systems can stop or send a payment.
- It does not determine intent. A mismatch is a mismatch. It does not establish that fraud occurred, who was responsible, or whether anyone acted in bad faith. A typo and a criminal alteration look identical to a comparison.
- It does not guarantee fraud prevention. No control does. A payment altered before authorization is sealed, or altered in a channel outside the sealed workflow, will not be caught by comparing against that seal.
- It does not guarantee any regulatory outcome. This is evidence infrastructure, not legal advice, not a compliance determination, and not a regulatory approval. It does not by itself satisfy any rule, examination, or reporting obligation.
- It does not replace your existing controls. Authentication, risk analytics and case management all still do jobs this does not do. Removing any of them because of this would be a mistake.
- It does not vouch for the truth of the underlying facts. A sealed record proves what was captured, when, and that it has not changed since. It does not prove that the submitted details were correct in the first place.
- Timestamp witnesses are not certifications. Independent timestamp authorities and external anchoring establish that a hash existed at a point in time. They are not a government certification, a regulatory approval, or an endorsement.
Where we honestly are
You are going to find this out in diligence, so you should hear it from us first.
- We are not deployed in a bank. CapchaCloud has no bank customers in production today and no case studies to show you. Anyone claiming otherwise on our behalf is wrong.
- Institutional security review is ahead of us, not behind us. Your security assessment, vendor onboarding, legal review, and production approvals all remain to be done, and we expect them to be demanding.
- We hold no certification of our own. CapchaCloud is pre-certification for its own SOC 2 Type II report and holds no ISO 27001 or PCI certification. We rely on our infrastructure and payment subprocessors' compliance postures plus our own documented safeguards. We do not display badges we have not earned. Current status: Trust Center.
- The verification gate is built and running, and it is version one. Today the independent re-statement compares recipient name and account number. Amount and other fields are sealed at authorization, so divergence from the sealed record is evident, but they are not independently re-stated by the confirming party in the current version.
- We are a small team. CapchaCloud is operated by ServiceAutomations.ai. That is a real consideration for a bank's concentration and continuity assessment, and we would rather discuss it early than have it surface late.
Questions your diligence team will ask
Does CapchaCloud move, hold, or custody funds?
No. CapchaCloud never executes, custodies, holds, or moves funds. Every response from the verification gate carries a standing advisory note saying so. The bank's own systems move money; CapchaCloud only compares details and seals a record of the comparison.
Does CapchaCloud block payments?
No. CapchaCloud records a verdict on its own record — matched, mismatched, or awaiting confirmation — and returns it with a sealed receipt. Nothing in CapchaCloud reaches into a payment rail. An authorized bank employee decides whether to release, hold, or investigate.
Is this a fraud score or a machine-learning model?
No. There is no score, no model, and no classifier anywhere in this path. It is a deterministic comparison: the details a confirming party independently re-states are normalized and compared against the details sealed at authorization. The result is a match or a mismatch, plus evidence of the check.
Can we verify an Evidence Record without trusting CapchaCloud?
Yes. That is the point. Download the standalone verifier and the evidence bundle and run the check offline, on your own hardware, with no network calls to CapchaCloud. The verifier re-derives the record hash, checks the signatures, and checks the record's inclusion in the append-only ledger.
Can we evaluate this without touching production?
Yes, two ways. A retrospective audit runs entirely on a closed, historical window of already-resolved payment activity and touches no live workflow. Shadow mode runs the comparison alongside your existing process without gating, blocking, or altering anything in the live path.
Does CapchaCloud hold a SOC 2 or ISO 27001 certification?
No. CapchaCloud is pre-certification for its own SOC 2 Type II report and holds no ISO 27001 or PCI certification of its own. We publish that plainly rather than implying otherwise. Current posture is documented in the Trust Center.
← Home · Evidence methodology · Crypto inventory · Trust Center · Explorer · Support