← all systems · see it in The System
Business email compromise
A class of cybercrime that uses email fraud to attack organisations (src:wp:email-spoofing): an offender impersonates an employee, supplier or other trusted business party, through a taken-over account, a spoofed address or a look-alike identity, to obtain funds, account numbers or sensitive information (src:cross-gillett-2020); in the FBI's words a scam targeting businesses or individuals who regularly make wire transfers, carried out by compromising email and other channels through social engineering or computer intrusion to make unauthorised transfers (src:ic3-2025). Walked from the defender's and the researcher's side: its kinds, losses and their limits, the people and institutions on each side, the defences and how well they are measured to work, and why the field says it works.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | by 2013 (IC3 statistics from October 2013; Google and Facebook defrauded 2013-2015) |
| born precision | year |
| boundary | Inside: the scheme from reconnaissance to the return: the information gathered, the borrowed identity, the tailored request, the recipient's decision as the fraud meets it, the transfer and its laundering through mules, and the groups' division of labour. Outside: the organisation and its employee (who are deceived), the banks, email providers and authentication, law enforcement and IC3, and the training and controls that answer it; they are partners, and their acts are recorded as channels, loops and observations. |
| nested level | Runs on organisations' payment processes and on people's judgement (Human L5-L7), carried by email; a variant of phishing and of social engineering; a class of cybercrime run by offender groups. |
| scale | technological |
Measures (33) published science
| BEC complaints and reported losses (IC3) | 2025: 24,768 and $3,046,598,558; 2024: 21,442 and $2,770,151,146; 2023: 21,489 and $2,946,830,270 complaints; US$ |
| BEC losses and complaints reported to IC3, 2014-2018 | 2014: $226,000,000 (2,417); 2015: $246,226,016 (7,837); 2016: $360,513,961 (12,005); 2017: $676,151,185 (15,690); 2018: $1,297,803,489 (20,373) US$ (complaints) |
| Exposed BEC losses, October 2013 to December 2022 | 277,918 incidents; $50,871,249,501 (domestic and international, from complaints, law enforcement and bank filings) incidents; US$ |
| US and international BEC exposed losses (actual plus attempted), June 2016 to July 2019 (FBI) | $26,201,775,589 over 166,349 incidents (Cross and Gillett give it as 'over 26 billion' lost) billion US$ |
| Net direct losses as a share of gross BEC losses (estimate) | 50-75 % |
| Countries reporting BEC; countries receiving transfers | 177; over 140 countries |
| Real estate BEC growth 2020-2022 | reports +27%; losses +72% % change |
| Financial Fraud Kill Chain, 2025 (all crime types; no BEC-only figure published) | 3,900 incidents; $1,163,919,846 attempted theft; $679,013,183 frozen; 58% success (domestic 3,574 incidents, $507,042,623 frozen; international 326, $171,970,560) incidents; US$; % |
| Kill chain for critical-infrastructure victims, 2025 (all crime types; no BEC-only figure published) | 655 incidents; $261,451,001 reported; $146,561,094 frozen; 56% incidents; US$; % |
| BEC complaints from people aged 60 and over (IC3 2025) | 4,566 complaints; $568,048,472 complaints; US$ |
| BEC losses involving AI reported to IC3 (2025) | over 30 million US$ |
| Objectives of sampled BEC attacks | wire transfer 46.9%; click link 40.1%; establish rapport 12.2%; steal PII 0.8% (3,000 attacks) |
| Impersonated roles in sampled BEC attacks | CEO 42.9%; CFO 2.2%; other C-level 4.5%; finance/HR 2.2%; other 48.1% (recipients: CEO 2.2%, CFO 16.9%, C-level 10.2%, finance/HR 16.9%, other 53.7%) |
| BEC share of all email (one vendor's customers) | less than 1 in 50,000 emails |
| BEC-Guard precision and false-positive rate | 98.2%; 1 in 5.3 million emails %; rate |
| CAPE detector precision over two years in production | above 80 % |
| SPF and DMARC adoption among the top million domains (2017) | SPF 45.0%; DMARC 4.6% (2015 scans as cited: SPF 40%, DMARC 1%) |
| Email providers showing a security cue on unverified mail | 9 of 35 providers |
| Click rate on spoofed mail with and without a security cue | 48.9% without, 37.2% with (role play, significant); real-phishing difference not significant |
| Phishing training effect on clicks (reproduction) | no significant effect (p = 0.450; reports p = 0.417); click rate 7.0% easy to 15.0% hard lures p; % |
| Phishing training trial size (Ho et al.) | over 19,500 employees; 10 simulated campaigns over 8 months |
| Operation Wire Wire (2018) | 74 arrests; about $2.4 million seized; about $14 million recovered arrests; US$ |
| Operation reWired (2019) | 281 arrests (167 in Nigeria); nearly $3.7 million seized; about $118 million recovered arrests; US$ |
| UK authorised push payment fraud (2022) | £485.2 million in 207,372 incidents (200,643 on personal accounts) GBP; incidents |
| Free webmail providers used for BEC | Google 68%; Microsoft 17% % of webmail-sent attacks (as cited) |
| BEC groups profiled; with their own domains; with mail servers | 20; 16; 13 groups |
| Price of a commodity remote-access tool used by BEC groups | 12-69 US$ |
| Targeted against mass campaign yield (as cited) | 10 times the money from 1,000 times fewer emails ratio |
| Reconnaissance lead time before a BEC | weeks or months time |
| IC3 complaint intake | almost 3,000 complaints per day (all crime types) |
| Reed warbler nests parasitised by cuckoos, Wicken Fen | 24% (1985) to 1% (2012); mobbing 90% to 38%; rejecting non-mimetic eggs 61% to 11% |
| Payroll diversion BEC complaints, January 2018 to June 2019 | 1,053 complaints; $8,323,354 reported; average $7,904 complaints; US$ |
| Lain et al. field study size | more than 14,000 employees over 15 months employees; months |
Open contentions (5) open contention
Where sources or readings disagree. Both stay on the map.
- C-3907 Is business email compromise a kind of spear phishing, the other way round, or the same thing as 'whaling'?
booked: The map keeps the competing readings; none is ruled.
the other reading: BEC is a kind of spear phishing (Al-Musib et al., as cited); spear phishing is a kind of BEC (GreatHorn, as cited); BEC equals whaling (Cidon et al., as cited); BEC is only text-based attacks with no link or attachment (Agari and others, as cited); drop both terms for 'targeted attacks' with qualifiers (Wassermann et al.) - C-3908 What share of business-email scam messages carry a link: about 40%, or about 3%?
booked: The map keeps the competing readings; none is ruled.
the other reading: the difference is definitional: one counts credential phishing by impersonation as BEC, the other counts text requests; only 41% of BECs involved phishing (DBIR 2022, as cited by Papathanasiou et al.): a third-hand industry figure whose meaning depends on the definition - C-3909 How much money does business email compromise really take: billions reported, tens of billions 'exposed', or far more unreported?
booked: The map keeps the competing readings; none is ruled.
the other reading: reported losses overstate (gross, not net of recoveries); reported losses understate (under-reporting, reputational silence); exposed losses overstate realised loss by design, because they count attempts (IC3 2019 PSA definition); a flat or rising complaint series reads reporting behaviour as much as offending (IC3 2019 PSA) - C-3910 Are business-email scam groups mostly based in Nigeria, or is that where the investigators have looked?
booked: The map keeps the competing readings; none is ruled.
the other reading: early law enforcement suspected Asia-based organised crime (funds traced to China, Japan, South Korea); the share reflects which groups vendors track, not the population of offenders; which offenders are labelled is itself selective: in Nigeria, tweet responses suggest the EFCC looks downwards for culprits (Lazarus, Button and Adogame 2022); Nigerian criminology reads 'Yahoo Boys' fraud as culture in action (Lazarus 2018) - C-3911 Does security awareness training stop staff falling for scam emails, or can it make things worse?
booked: The map keeps the competing readings; none is ruled.
the other reading: training reduces BEC losses; training as deployed has no practical effect on clicks; process controls (call-back, sign-off) carry the defence
Cross-scale patterns seen here (12) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Business email compromise. Each links to every other system that shows it.
- same role, different builds (structural)
One job, getting a carer or payer to spend its resources on an impostor by copying the signature of one it already serves, done by unrelated builds: a BEC request in a CEO's or supplier's name diverting a company's payment; a cuckoo finch egg mimicking the host's eggs diverting parental care; a reed - trade one property for another (structural)
Catching more fraud is paid for in blocked legitimate mail: organisations do not enforce DMARC because it blocks legitimate email; filters need a balance between blocking fraud and blocking business; a detector tuned for precision (98.2%) loses recall (its impersonation stage alone runs at 11.7% pre - Umwelt: each system carves its own slice of one shared medium (structural)
One message, read in different slices: the receiving server sees SPF/DKIM/DMARC results the user interface mostly does not show (9 of 35 providers show a cue); DMARC reads headers only; the employee sees the display name, often not the address; the fraud sees the organisation only through public pag - who opens the door (structural)
The door the money leaves by is opened from inside, ahead of time: the organisation built a payment process that acts on email instructions, and the fraud only addresses that port. Its listeners (call-back, sign-off, payee name check) were installed by the organisation and the banks. - causal ceiling (partial)
No filter can tell a request sent from a genuine, taken-over account from a real one; only the recipient can spot a fake, and the person whose address is spoofed cannot know unless DMARC reporting is on. The limit is set by the mail system's structure, not the filter's means. - red-queen (partial)
Detectors are retrained as requests change (gift cards slipped past); people learned to spot machine-translated text and the groups moved to AI-written text; IC3-reported losses stayed between $2.77 and $3.05 billion a year from 2023 to 2025 while both sides changed. The stalemate is read from three - same shape, opposite sign (partial)
One move, moving the conversation to a second channel, with its sign reversed: the defence calls back on a known number to confirm the request; the fraud moves the target to a lawyer or phone number it controls to keep the request from being confirmed. - same toolkit, independent builds (partial)
The two-step CEO-then-attorney scheme is run by a Russian group (Cosmic Lynx) and an Israel-based group, from the same parts (email, a CEO's name, a lawyer's authority, an acquisition story); the Nigerian groups may share connections, so independence is shown only between those two. - templating (partial)
The same off-the-shelf malware (Agent Tesla, NETWIRE, Dark Comet) is used by several groups, and the same kinds of request recur across groups. - The carry (partial)
The wait face: recall works only while the money is still in the first account; once it is carried to a second hop the kill chain must chase it. The lookahead face: verifying before paying (a call-back, Confirmation of Payee) settles the question before the money moves. - The Payment to Entropy (partial)
Value form: each hand-off of the stolen money pays a toll (mules' commission) and some is caught at each hop (banks stop or revert transfers: net losses likely 50-75% of gross; 58% of attempted theft frozen across 2025 kill-chain cases). No per-hop efficiency is sourced; much of the 'loss' is a hand - lightning (surface)
Trust built up over a relationship is discharged in one wire, after which the organisation is not the same: FACC lost €42 million and fired its CFO and CEO; lawsuits against employees followed elsewhere.
Valence: what it trades, and with whom (12) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Human (Homo sapiens) via sys:electromagnetic-field · Business email compromise gain
- Large language model via sys:written-record · Business email compromise gain
- Commercial banks (creators of most money) via sys:payment-system · Business email compromise mixed
- Police via sys:human-society · Business email compromise loss
- Authentication via sys:email · Business email compromise loss
- Threat actor via sys:internet · Business email compromise gain
- Phishing via sys:internet · Business email compromise gain
- Organization via sys:email · Business email compromise gain
- Money mule via sys:payment-system · Business email compromise gain
- Social engineering via sys:human-society · Business email compromise gain
- Security awareness via sys:mental-representation · Business email compromise even
- Romance scam via sys:human-society · Business email compromise gain
Partners and parts published science
Built from / builds (5)
Sits inside / holds (3)
expressed by (13)
- Quality Improvement/Refinement
- Sensing/Scanning
- Mapping/Modeling
- Evade Detection
- Overcome Barrier
- Speed
- Precision Attack
- Ranged Attack
- Subversion & Co-optation
- Prepositioning & Latency
- Leverage Host/Amplification
- Divide & Isolate
- Deception
floor filled by (15)
- Human (Homo sapiens)
- Internet
- Large language model
- Commercial banks (creators of most money)
- Domain Name System
- Electronic mail
- Malware
- Threat actor
- Phishing
- Organization
- Cybercrime and its underground economy
- Money mule
- Social engineering
- Account takeover
- Internet Crime Complaint Center (IC3)
represents (1)
Kinds (11)
- Phishing
- Social engineering
- CEO fraud (executive impersonation)
- Fake invoice and vendor email compromise
- Attorney impersonation
- Data theft (W-2 and employee data)
- Gift-card scam
- Commodity theft
- Email account compromise (EAC; individuals, often real estate)
- Payroll diversion
- Confidence trick
Notes published science
- The scam's most common words are not threats but office chores: 'urgent', 'request', 'important'. Researchers say it works less through fear or greed than through an employee's wish to do the job well for the boss, and one detector found fewer than one email in 50,000 is a BEC attack, with nothing malicious in it for a filter to catch (src:cross-gillett-2020, src:cidon-2019).
- When a company's money goes to a fraudster, the race is on: the FBI's 'Financial Fraud Kill Chain' asks the receiving bank to freeze the account before the money hops onward, and in 2025 it froze $679 million of $1.16 billion in attempted thefts across all the frauds it handled. Wire transfers are chosen because they are easy to send and hard to reverse (src:ic3-2025, src:papathanasiou-2023).
- Business email compromise and the cuckoo run the same trick: copy a trusted signature to divert someone else's care. The defences pair up too. Zitting cisticolas, whose eggs come in distinct 'signature' categories and who reject only eggs of another category, make fewer mistakes of both kinds than birds with smoothly varying eggs, much as a cryptographic signature or a payee-name check tests the message itself. Reed warblers reject more eggs after they see a cuckoo at the nest, a second cue from outside the egg, as a call-back to a number already on file is a second channel the impostor does not control (src:dixit-2026, src:thorogood-davies-2013-cuckoo).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
business email compromise as fraud researchers, criminologists and security economists describe it: the trusted business relationships it imitates, up to the money and data it diverts (reader's view)
- L1: The trusted relationships of business and the mail they travel by: executives and staff, suppliers, customers, lawyers, title companies; email (and phones and meeting apps) used to arrange payments; the organisation's public face
- L2: Reconnaissance: names, roles, relationships, processes, travel and writing style gathered from public sources, data breaches and compromised mailboxes, for weeks or months
- L3: The borrowed identity: a taken-over real account, a spoofed address, or a look-alike domain or free webmail account under a trusted name
- L4: The scheme: a tailored request (pay this transfer, change these bank details, buy gift cards, send W-2 forms) timed and worded for the person who can act
- L5: The recipient's decision inside the organisation's payment process: comply, or verify by another channel; runs on the employee (Human) and the organisation's controls
- L6: The transfer: funds wired through banks to accounts the fraudsters control, moved on by money mules and second hops, often abroad; recall and freezing contest it here
- L7: The return: financial gain (and data for further fraud) for loosely organised groups that divide labour like firms
the boardroom (shared floors)
- L4: The scheme: a request built from what reconnaissance found, sent under a borrowed identity to the person who can act; tailored, rare and carrying no malicious payload
- L5: The recipient's decision inside the payment process: authority, urgency and routine against verification and sign-off. Runs on the employee (Human's board) and the organisation's decision floor
- L6: The transfer and its contest: money wired to accounts the fraudsters control and moved on by mules, against recall, freezing and investigation; since 2022 also to cryptocurrency exchanges, which changes the race the freeze has to run
- L7: The return: financial gain and data for other fraud, for offender groups with a division of labour like a firm's
way in: A compromised mailbox
- L1: A legitimate account of an executive or supplier, taken over through malware or phishing
- L2: Access gained, often weeks or months before the fraud
- L3: Conversations, contacts and real invoices read; genuine documents altered with new bank details
way in: Authority: the boss's name
- L1: The organisation's hierarchy of executives and staff
- L2: A request in the name of the CEO or another senior person; society trains people not to question authority
- L3: The wish to do the job well and please management
- L4: Compliance with what looks like a direct order
way in: Controls: verification, sign-off and training
- L1: Governance processes to validate all payments consistently
- L2: Verification by a second channel (a phone call to a known number) or two-factor authentication for changes to account details
- L3: Several people sign off large transactions; email money requests always queried by other means
- L4: Awareness training (measured effects small or null in phishing trials) and goals that can pull against compliance; reporting: employees with a report button as a collective detector (shown for link phishing)
way in: Detection met on the way: authentication and filters
- L1: SMTP carries no built-in sender authentication
- L2: SPF, DKIM and DMARC check the sending domain; adoption low (DMARC 4.6% of top domains in 2017); headers only
- L3: Filters and detectors: bulk filters miss tailored mail; detectors that learn each organisation's history reach high precision
way in: Law enforcement
- L1: Victims' complaints: IC3 averages almost 3,000 a day
- L2: Analysts assign crime type and adjusted loss; figures are a point-in-time assessment
- L3: Investigation follows the money: bank secrecy lifted, email providers asked for records
- L4: Coordinated operations: Wire Wire (2018), reWired (2019)
- L5: International cooperation: European Investigation Orders, Joint Investigation Teams, mutual legal assistance
way in: Money mules
- L1: People reached by romance and work-at-home scams
- L2: Recruited, at times unwittingly, as 'financial agents'
- L3: Their accounts receive the victim's money for a small commission; across online fraud generally, online and international accounts and crypto wallets are favoured, and mules are recruited for identity cards to open them (Bekkers)
- L4: Funds moved on, often abroad (Hong Kong and China the main 2022 destinations; the UK an intermediary stop, then Mexico and Singapore), or straight to a cryptocurrency exchange or its custodial bank
- L5: The paper trail broken; lost funds hard to trace
way in: Public organisational information
- L1: Organisational information freely available online; social networks show relationships between employees
- L2: Gathered as open-source intelligence: employee names, email addresses, roles (the three top-scoring reconnaissance techniques across twenty groups); lead-generation software
- L3: A profile of the target: culture, relationships, how staff talk to each other, travel movements
way in: Recall and freezing
- L1: The victim discovers the transfer; time is of the essence
- L2: The victim's bank asked to recall the funds; a complaint filed with IC3
- L3: The IC3 Recovery Asset Team starts the Financial Fraud Kill Chain with banks and field offices
- L4: The recipient bank freezes the account if the money is still there
- L5: Second hops and international transfers followed (FinCEN Rapid Response Team, legal attachés, foreign police)
way in: Routine: the expected invoice
- L1: Suppliers in other time zones; invoices commonly settled by email
- L2: An invoice within the bounds of usual expenditure, with the hallmarks of legitimate business; or a spoofed 'employee' asking HR to move a salary deposit (payroll diversion), a junior and mundane request
- L3: A genuine invoice with altered bank details, sent in the real thread; in real estate, a change of payment type (cheque to wire) or of receiving account
- L4: Hard to tell from real mail, since payment is a normal topic with vendors
way in: Sender identity: addresses, domains and accounts
- L1: Domain names and free webmail accounts (Google 68%, Microsoft 17% of webmail BEC, as cited)
- L2: A look-alike domain registered (sixteen of twenty groups registered their own), or a trusted address spoofed
- L3: A From line, display name and reply-to that read as the trusted party
way in: Sight: the message on a screen
- L1: Light from a screen, often a phone
- L2: The eye reads what the mail client shows
- L3: Many readers do not display the sender's address; display name and reply-to differ unseen
- L4: Look-alike letters pass at a glance; few providers add a warning cue, which helps in role play and less in real tests; warnings on emails effective in a 15-month field study
way in: The bank's checks on the payment
- L1: The payment system: wire transfers, easy to make and hard to reverse
- L2: Confirmation of Payee (UK): the name on the receiving account checked when a new payee is set up
- L3: Fraud checks in the payment journey; customers can be coached by fraudsters past them
- L4: Who bears the loss: reimbursement codes cover consumers and micro-enterprises, not larger businesses
- L5: Banks detect and stop or revert some BEC transfers
way in: Tools: commodity malware, lead software, language models
- L1: Hacking forums and online markets; legitimate marketing lead-generation software
- L2: Off-the-shelf remote-access tools and keyloggers bought cheaply (one for $12 to $69)
- L3: Fluent text in many languages from AI language models, after people learned to spot machine translation
way in: Urgency: time pressure
- L1: The business clock: requests near the end of the business day
- L2: Words such as 'urgent', 'request', 'important'
- L3: Under time pressure, a decision strategy with less reasoning
- L4: Less thinking through, less questioning
way out: Data released
- L1: The data used for identity theft and false tax returns
- L5: HR or finance staff send W-2 forms or employee lists
way out: Funds moved on
- L1: Funds abroad or converted at a cryptocurrency exchange, the trail broken
- L3: Mules move the funds on for a commission
- L6: Money flow managed by a dedicated role
way out: The payment made
- L1: Funds arrive in an account the fraudsters control
- L3: The organisation's bank sends the wire
- L5: The employee approves the payment or the change of bank details
way out: The request sent
- L1: The message in the inbox of the person who can pay or release data
- L3: Sent under the borrowed identity, from the taken-over account or a look-alike address
- L4: The tailored request composed
way out: The victim's report
- L1: The report reaches the bank and IC3
- L2: A recall requested from the bank and a complaint filed
- L5: The fraud noticed (irregularities, a missing deposit, a supplier asking for payment)