{
 "slug": "deepfake-wire-fraud-ucc-article-4a",
 "topic_id": "TOPIC-001",
 "cluster": "B2B Wire, AP & Treasury Payment Release",
 "tier": "Tier A",
 "title": "The deepfake wire trap: why out-of-band callbacks fail UCC Article 4A",
 "summary": "Treasury policy mandates a phone callback to the CFO before releasing a large wire. Voice and video cloning defeat that control, and the commercial reasonableness standard in UCC §4A-201 was written for a world where a voice was evidence.",
 "lede": "Article 4A of the Uniform Commercial Code allocates loss on a fraudulent payment order according to whether the bank's security procedure was commercially reasonable and followed in good faith. That test was written when a telephone callback was a strong control. It is now the weakest link in the chain.",
 "date": "2026-05-09",
 "category": "Compliance",
 "author_id": "constance-ibe-whitmore",
 "tags": [
  "UCC Article 4A",
  "wire fraud",
  "deepfake",
  "treasury",
  "callback verification",
  "commercial reasonableness"
 ],
 "image_title": "Deepfake Wire Trap",
 "schema": "Article",
 "key_takeaways": [
  "UCC §4A-201 through §4A-203 allocate loss based on whether a commercially reasonable security procedure was agreed and followed. The standard is relative to the state of practice, and practice has changed.",
  "A telephone callback binds nothing to the payment order. Voice is an analog channel with no cryptographic relationship to the amount, the beneficiary or the account.",
  "A signed payment order statement moves the control from recognising a voice to producing a signature, which shifts the evidentiary position materially."
 ],
 "body": [
  {
   "type": "h2",
   "text": "What Article 4A actually allocates"
  },
  {
   "type": "diagram",
   "kind": "chain",
   "alt": "Voice was evidence when the standard was written",
   "caption": "The procedure did not change. What a voice proves did.",
   "nodes": [
    {
     "label": "Policy: call the CFO",
     "sub": "before release",
     "note": "the control"
    },
    {
     "label": "Voice cloned",
     "sub": "seconds of audio",
     "note": "commodity",
     "bad": true
    },
    {
     "label": "Video call, several colleagues",
     "sub": "all synthetic",
     "note": "Arup, 2024",
     "bad": true
    },
    {
     "label": "Release approved",
     "sub": "procedure followed",
     "note": "loss",
     "bad": true
    }
   ]
  },
  {
   "type": "p",
   "html": "Under UCC §4A-202, a payment order received by a receiving bank is effective as the order of the customer if the bank and the customer agreed that the authenticity of payment orders would be verified pursuant to a security procedure, and the bank accepted the order in good faith and in compliance with that procedure."
  },
  {
   "type": "p",
   "html": "§4A-201 defines a security procedure as a procedure established by agreement for verifying that a payment order is that of the customer, or for detecting error. §4A-202(c) provides that commercial reasonableness is a question of law, determined by reference to the wishes of the customer expressed to the bank, the circumstances of the customer known to the bank, alternative procedures offered, and procedures in general use by similarly situated customers and banks."
  },
  {
   "type": "p",
   "html": "Read that last clause carefully. Commercial reasonableness is measured against <em>procedures in general use</em>, which means the standard moves as practice moves. A procedure that was reasonable in 2015 is not automatically reasonable now."
  },
  {
   "type": "h2",
   "text": "What a callback verifies"
  },
  {
   "type": "p",
   "html": "A treasury callback establishes, at best, that a voice on a telephone line associated with a number on file agreed to a payment. Decompose what that does and does not establish."
  },
  {
   "type": "table",
   "head": [
    "Element",
    "Verified by the callback?"
   ],
   "rows": [
    [
     "A call was placed to a number on file",
     "Yes, if the number was not forwarded"
    ],
    [
     "Someone answered",
     "Yes"
    ],
    [
     "The voice resembled the authorised person",
     "Yes, to the caller's ear"
    ],
    [
     "The voice <em>was</em> the authorised person",
     "<strong style=\"font-weight:600\">No</strong>"
    ],
    [
     "The amount discussed matches the payment order",
     "Only as spoken and recalled"
    ],
    [
     "The beneficiary account matches the payment order",
     "Rarely read aloud in full"
    ],
    [
     "Any of the above is demonstrable afterwards",
     "<strong style=\"font-weight:600\">No</strong>"
    ]
   ]
  },
  {
   "type": "p",
   "html": "The last three rows are where the control fails as evidence even when it succeeds as a control. A callback log records that a call occurred; it does not record what was agreed."
  },
  {
   "type": "h2",
   "text": "Why generative voice changed the analysis"
  },
  {
   "type": "p",
   "html": "Voice cloning moved from research demonstration to commodity tooling over the past several years. Corporate executives are among the most exposed targets because their voices are publicly available in earnings calls, conference recordings and media appearances."
  },
  {
   "type": "p",
   "html": "The FBI's Internet Crime Complaint Center has reported business email compromise as one of the highest-loss cybercrime categories, with a substantial share of losses moving by wire or ACH — payment rails where recovery windows are short. Publicly reported cases have involved video conferences in which participants were synthetic."
  },
  {
   "type": "p",
   "html": "None of that requires the attacker to break any cryptography, because none is in use. The callback is an analog channel evaluated by human perception."
  },
  {
   "type": "h2",
   "text": "The commercial reasonableness argument"
  },
  {
   "type": "p",
   "html": "Here is the argument a general counsel should be prepared to encounter, framed from the bank's side."
  },
  {
   "type": "p",
   "html": "If procedures in general use among similarly situated customers include cryptographic verification of payment orders, and a customer declined an offered procedure of that kind in favour of a telephone callback, §4A-202(c) contemplates that the customer may be bound by the procedure it chose. The provision expressly addresses the case where a customer refuses a security procedure offered by the bank."
  },
  {
   "type": "p",
   "html": "That is not a prediction about how any court will rule, and this article gives no legal advice. It is an observation that the statute makes the customer's choice of procedure relevant, and that the set of available procedures has changed."
  },
  {
   "type": "h2",
   "text": "What a bound payment order looks like"
  },
  {
   "type": "p",
   "html": "The control that addresses this does not try to detect the fake. It requires something the fake cannot produce."
  },
  {
   "type": "code",
   "text": "{\n  \"type\": \"manav-stmt/1\",\n  \"action\": \"authorise_payment_order\",\n  \"render\": [\n    \"Amount: [currency] [amount in full]\",\n    \"Beneficiary: [full legal name]\",\n    \"Account: [full account number]\",\n    \"Routing / SWIFT: [full code]\",\n    \"Bank: [beneficiary bank, country]\",\n    \"Reference: [invoice or contract]\",\n    \"Value date: [date]\"\n  ],\n  \"authoriser\": \"[credential assertion, user verification required]\"\n}"
  },
  {
   "type": "p",
   "html": "The details are rendered in full and unmasked, because masking hides exactly the characters an attacker alters. The signature covers a canonical representation of that content, so a payment order altered after authorisation fails verification."
  },
  {
   "type": "h2",
   "text": "What this changes for the treasury team"
  },
  {
   "type": "p",
   "html": "Less than it appears. The approver still reviews the payment. The difference is that their confirmation is a device gesture producing a signed record rather than a telephone conversation producing a log entry."
  },
  {
   "type": "p",
   "html": "What changes materially is the position afterwards. In a dispute with a bank, an insurer or a counterparty, the company holds a verifiable record of exactly what its authorised officer approved — which is the artefact that currently does not exist anywhere in the process."
  },
  {
   "type": "h2",
   "text": "The honest limits"
  },
  {
   "type": "p",
   "html": "This stops an unauthorised payment order. It does not stop a payment that a genuine, authorised person was deceived into approving with full knowledge of the details — the invoice was fraudulent but the approver read it, understood it, and approved it."
  },
  {
   "type": "p",
   "html": "For that case the control is a different one: rendering the beneficiary's payment history, flagging first-time payees, and requiring a second signature on new beneficiaries. Those are worth doing and they are a separate argument."
  },
  {
   "type": "h2",
   "text": "What survives, and what does not"
  },
  {
   "type": "table",
   "caption": "Factors against a real-time impersonation",
   "head": [
    "Factor",
    "Survives?"
   ],
   "rows": [
    [
     "Recognising a voice",
     "No"
    ],
    [
     "Recognising a face on video",
     "No"
    ],
    [
     "A shared secret spoken aloud",
     "No — relayed in real time"
    ],
    [
     "A code sent to a number on file",
     "Only if the number is not attacker-controlled"
    ],
    [
     "<strong style=\"font-weight:600\">A signature from an enrolled device</strong>",
     "<strong style=\"font-weight:600\">Yes — the caller cannot produce it</strong>"
    ]
   ]
  },
  {
   "type": "p",
   "html": "The Article 4A consequence is the part treasurers miss: a customer offered a stronger procedure who declined it is generally in a worse position than one who took it. As stronger procedures become available and adopted, the commercial reasonableness of a voice callback erodes without anyone changing your agreement."
  },
  {
   "type": "h2",
   "text": "Objections and honest limits"
  },
  {
   "type": "p",
   "html": "<strong style=\"font-weight:600\">“Our callback uses a number from the vendor record.”</strong> Better, and it still authenticates a channel rather than a person. A cloned voice answering the correct number passes."
  },
  {
   "type": "p",
   "html": "<strong style=\"font-weight:600\">“We would detect a deepfake.”</strong> Detection is a probabilistic contest against improving synthesis, on a call the attacker chose the timing of. It is the wrong side of the trade."
  }
 ],
 "faq": [
  {
   "q": "Does this article give legal advice on UCC Article 4A?",
   "a": "No. It describes the statutory framework and observes that commercial reasonableness is measured against procedures in general use. How any court would apply it to specific facts is a question for counsel."
  },
  {
   "q": "Should we stop doing callbacks?",
   "a": "No. Keep them for the conversation and the relationship. Stop treating them as the verification of record, and stop describing them as such in your control documentation."
  },
  {
   "q": "What about payments below the callback threshold?",
   "a": "Apply the signature above a value threshold computed from your own payment distribution, not from a round number. Most payment populations are heavily skewed."
  },
  {
   "q": "How does this affect our banking agreement?",
   "a": "Security procedures are agreed with the bank. Adding a customer-side control is worth raising with them, because banks increasingly price and document these arrangements."
  },
  {
   "q": "Is a phone callback worthless?",
   "a": "Not worthless, and no longer sufficient on its own for high-value releases. It authenticates a channel, and cloning defeats the voice."
  },
  {
   "q": "Why does declining a stronger procedure matter?",
   "a": "Under Article 4A a customer offered a procedure who declined it has generally accepted a different risk allocation."
  },
  {
   "q": "What replaces it?",
   "a": "A signature from a device enrolled to the approver, bound to the specific payment. The caller cannot produce it."
  }
 ],
 "sources": [
  {
   "t": "UCC Article 4A — Funds Transfers (Cornell LII)",
   "u": "https://www.law.cornell.edu/ucc/4A"
  },
  {
   "t": "FBI IC3 2025 Internet Crime Report",
   "u": "https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf"
  },
  {
   "t": "FCC — protecting consumers from SIM swap and port-out fraud",
   "u": "https://www.fcc.gov/sim-swap-port-out-fraud"
  },
  {
   "t": "RFC 8785 — JSON Canonicalization Scheme",
   "u": "https://www.rfc-editor.org/rfc/rfc8785"
  },
  {
   "t": "W3C Web Authentication Level 2",
   "u": "https://www.w3.org/TR/webauthn-2/"
  }
 ],
 "related": [
  {
   "slug": "erp-dual-authorization-flaw-sox-controls",
   "title": "The four-eyes illusion",
   "category": "Developer"
  },
  {
   "slug": "vendor-bank-account-change-fraud-controls",
   "title": "The vendor invoice scam",
   "category": "Compliance"
  },
  {
   "slug": "cyber-insurance-callback-warranty-wire-fraud-denials",
   "title": "The callback warranty trap",
   "category": "Compliance"
  },
  {
   "slug": "ucc-article-4a-corporate-wire-exposure-calculator",
   "title": "Modelling corporate wire exposure under UCC 4A",
   "category": "Compliance"
  }
 ],
 "image": "https://cdn.twc.sh/images/igcache/Deepfake%20Wire%20Trap/1200_630/blog.jpg",
 "wordcount": 1229,
 "url": "/blog/deepfake-wire-fraud-ucc-article-4a.html",
 "reading_time": "5 min read",
 "seo_title": "Why out-of-band callbacks fail UCC Article 4A",
 "meta_description": "Treasury policy mandates a phone callback before releasing a large wire. Voice and video cloning defeat that control at commodity cost.",
 "hub": {
  "slug": "topics/payment-release-authorization",
  "title": "Payment release authorization"
 },
 "answer": "It is increasingly hard to argue. UCC Article 4A judges a security procedure against the customer's circumstances and the alternatives available. Voice and video cloning are commodity, the Arup case showed a multi-person video call defeated, and a procedure whose only factor is recognising a voice is defending against a capability that no longer exists.",
 "answer_q": "Does a phone callback still satisfy a commercially reasonable security procedure?",
 "glossary": [
  {
   "term": "Commercially reasonable",
   "def": "The Article 4A standard, assessed against the customer's circumstances and the alternatives offered."
  },
  {
   "term": "Real-time impersonation",
   "def": "Synthesised voice or video generated live, so interaction does not defeat it."
  },
  {
   "term": "Out-of-band",
   "def": "Through a channel independent of the one carrying the request. Only as good as how the second channel was obtained."
  }
 ],
 "checklist": {
  "title": "Re-examining your callback procedure",
  "id": "reexamine",
  "desc": "Four questions.",
  "steps": [
   {
    "name": "What does the callback actually test?",
    "text": "Channel, voice, or a signature. Only the third resists cloning."
   },
   {
    "name": "Where did the number come from?",
    "text": "The pre-change record, or the request?"
   },
   {
    "name": "What has the bank offered that we declined?",
    "text": "And is the declination on file?"
   },
   {
    "name": "What do we produce afterwards?",
    "text": "A note, or an artefact showing what was confirmed."
   }
  ]
 },
 "cta": {
  "title": "Where this fits in Manav",
  "html": "Manav renders the payment or change in full, binds the approver's signature to that exact payload, and produces a receipt an insurer, an auditor or a court can verify without calling anyone.",
  "href": "../docs.html",
  "label": "See payment gating"
 }
}