AI-era threat
AI-powered phishing and business email compromise
Published
Criminals now use AI to write fluent, personalized phishing emails and payment requests, in any language and at volume. Business email compromise (BEC), where a scammer poses as a boss, vendor or title company to redirect a payment, drew $3.05 billion in reported losses to the FBI in 2025.1 Typos are no longer the tell; a request to move money or change bank details is. Confirm every one on a second channel you already trust.
What it is
Phishing is a message that tricks you into clicking a link, typing a password, or paying. Business email compromise is phishing aimed squarely at money: the criminal poses as an executive, vendor, lawyer or title company, by email, phone or video meeting, and asks for a payment or a change of bank details.1
AI doesn't change the trick; it removes the friction. Chat tools translate away the spelling mistakes that used to give scammers away and produce tailored variants in seconds.2 Researchers estimate that by April 2025 about 14% of BEC emails in one large dataset were machine-written.3
By the numbers
$3.05 billion1
Business email compromise losses reported to the FBI in 2025, from 24,768 complaints; only investment fraud cost more.
FBI Internet Crime Complaint Center (IC3), p.7 (complaint counts) and p.8 (losses)
$30 million+1
2025 BEC losses in complaints that mentioned AI. The FBI says chat tools and voice cloning can be used in BEC; it tags AI only when a complaint mentions it, so this is a floor.
FBI Internet Crime Complaint Center (IC3), p.39 and p.42; 'AI Related' defined p.61
54% vs 12%4
Click rate for fully AI-automated spear-phishing emails in a 101-person study, matching emails written by human experts (54%) and far above generic phishing (12%).
Heiding, Lermen, Kao, Schneier and Vishwanath (arXiv:2412.00586), Abstract
About 14%3
Share of BEC emails in a large email-security dataset estimated to be AI-written as of April 2025, alongside about 51% of spam.
ACM Internet Measurement Conference (IMC) 2025; Hao et al., Columbia University and University of Chicago with Barracuda Networks, Abstract
74%5
Organizations hit by BEC attempts or losses in 2025, per a survey of 465 corporate treasury professionals. Firms with treasury staff tend to be larger than a typical small office, so read it as a signal, not a small-business rate.
Association for Financial Professionals (AFP), underwritten by Truist, Key Findings, p.3
$679 million1
Money the FBI's Recovery Asset Team froze in 2025 through the Financial Fraud Kill Chain, 58% of the $1.16 billion at stake across 3,900 incidents. Fast reporting is what makes a freeze possible.
FBI Internet Crime Complaint Center (IC3), p.17
48,0641
Phishing and spoofing complaints to the FBI from people 60 and older in 2025, about double the 23,252 filed in 2024.
FBI Internet Crime Complaint Center (IC3), p.47, three-year comparison for 60+
What it looks like
How this plays out, based on documented cases.
The vendor thread that switches banks
Your bookkeeper is trading emails with a regular supplier about an overdue invoice. A new reply lands in the same thread, friendly and fluent, quoting the invoice number and attaching 'updated' bank details because the supplier changed banks. In a 2026 survey of finance teams, criminals most often posed as vendors or executives asking to change payment instructions, and 39% of organizations had genuine emails intercepted by fraudsters.5 Nothing in the message looks wrong. The only reliable tell is the request itself: new bank details, arriving by email.
The owner who can't talk right now
Late on a Friday, an email that looks like it's from the owner asks you to pay a new consultant before close of business and keep it quiet until a deal is announced. When you hesitate, a voice note arrives that sounds just like them. The FBI says chat tools can produce official-sounding emails that mimic a chief executive, and cloned voices can be used to request wire payments.1 Treasury's Financial Crimes Enforcement Network (FinCEN) and Europol both report criminals using AI to impersonate executives and make this kind of CEO fraud more convincing.67
The closing-day wire
You're days from closing on a house when an email from the 'title company' arrives with wiring instructions for the down payment. In a case the FBI worked in March 2025, a senior in Missouri received a compromised title-company email directing a wire of more than $1.3 million to a criminal's account.1 After the complaint came in, the FBI's Recovery Asset Team had the receiving bank freeze the account. In an August 2025 case, a couple closing on a home wired over $449,000 after an email impersonating their attorneys, and after they filed with the FBI the receiving bank confirmed the full amount was still on hold.1
The 'security alert' text
A text that looks like it's from your bank, Amazon or Apple warns of a suspicious charge and gives you a link or a number to call. Among older adults who reported losing money to text scams in 2025, 31% described texts impersonating well-known businesses, often dressed up as security alerts.8 The link leads to a convincing login page, or the 'fraud team' on the phone tells you to move your savings somewhere 'safe'.8 AI makes these messages fluent, and the FBI has tracked campaigns that pair texts with AI-generated voice messages before sending a link that steals passwords.29
How it gets in
- A compromised mailbox: once criminals are inside your email or a vendor's, they can read real threads and reply inside them, from the right address.5
- Lookalike addresses and display names: a domain one letter off, or a familiar name over an unfamiliar address, which is especially hard to spot on a phone.10
- Fake login pages: a link captures your password, and without multi-factor authentication (MFA) the account is theirs.9
- Research at machine speed: AI tools can gather public details about you and your company and write to them; in one study an automated tool built accurate profiles of targets 88% of the time.4
- Fluent in any language: translation tools remove the grammar and spelling errors people were taught to look for.2
- Channel-hopping: the scam can start in email and move to a call, a text or a video meeting, where a cloned voice or face does the persuading.16
- Organized tooling: criminal groups use AI to draft phishing in several languages and to script call-center conversations.117
What stops it
Prevention: the habits and settings that break this attack.
Verify every money request on a second channel
Before you pay, add a payee or change bank details, call the person back on a number you already had, never one from the message. In a 2026 survey, finance teams rated this callback the most effective BEC control they use: 63% of organizations using it called it very effective.5
Require two people for large or unusual payments
BusinessAbove a dollar threshold you set, and for every new payee or bank-detail change, one person prepares the payment and a second approves it after their own callback. Ask your bank to enforce dual approval in its online portal so the rule doesn't depend on memory.
Turn on multi-factor authentication for email
MFA stops a stolen password from opening your mailbox, which is how many BEC schemes get inside real threads. Prefer phishing-resistant options such as passkeys or security keys, which FinCEN recommends to banks facing deepfake fraud.6
Publish email authentication for your domain
BusinessSender Policy Framework (SPF), DomainKeys Identified Mail (DKIM) and Domain-based Message Authentication, Reporting and Conformance (DMARC) tell receiving mail servers who may send as your domain. They make your exact address much harder to spoof, though they don't stop lookalike domains or a hijacked real account.
Train for the request, not the typo
Retire 'look for spelling mistakes' as your main tip, because AI removed most of them.2 Teach one pattern instead: money, plus a change, plus urgency or secrecy, means stop and verify. In one study, AI-written spear-phishing was clicked as often as emails crafted by human experts.4
Let a password manager fill in your logins
A password manager offers your saved login only on the real site, so a lookalike page gets nothing. That quiet refusal is often the first sign a link is fake.
Make reporting fast and blame-free
BusinessGive staff one easy way to flag a suspicious message, such as a report button or a shared address, and thank the people who use it. Early reports let you warn others and catch a compromised account before money moves.
Go to the source yourself
HomeFor a bank, store or delivery alert, don't use the link or number in the message. Open the company's app, type its address yourself, or call the number on the back of your card.2
Confirm home-closing and large wires by phone
HomeWhen you buy a home or send any large transfer, get the title company's or lawyer's number from your contract or their official website at the start, and call it to confirm wire instructions before you send. Treat any emailed change to those instructions as fraud until that call says otherwise.
Response, not prevention
If it already happened
Work through these in order. Moving quickly limits the damage, and there is no shame in reporting it.
Call your bank's fraud line immediately
Use the number on the bank's website or your card, not one from any email. Say the payment was fraud, ask for an immediate recall of the wire or transfer, ask your bank to send the receiving bank a hold harmless letter (the paperwork that lets that bank return the money), and write down the reference number.1
File a complaint at ic3.gov, whatever the amount
The Internet Crime Complaint Center (IC3) is the FBI's reporting site. Include full transaction details: date, amount, sending and receiving banks, account numbers, and the messages involved. Its Recovery Asset Team uses these reports to ask banks to freeze money, and in 2025 it froze $679 million across 3,900 incidents.1
If the scam reached you at home, also report it to the Federal Trade Commission (FTC) at ReportFraud.ftc.gov.12
Secure the email account involved
From a clean device, change the password, sign out every session, delete any forwarding rules or filters you didn't create, and turn on MFA. Tell your IT provider: a mailbox that sent or received the fake request may still be in the attacker's hands.
Warn the real person or company
Call the real vendor, executive, title company or lawyer on a number you already trust. Their account may be the one that was taken over, and their other customers could be next.
Preserve the evidence
Keep the emails with full headers, texts, call logs, invoices and bank confirmations. Don't delete anything, even the parts that feel embarrassing; investigators and insurers need the whole trail.
Run your incident playbook and call your insurer
BusinessOpen your incident response plan, notify your cyber insurer if you have one, and log every action with a time. Then fix the payment step that let the request through.
For businesses with an incident plan
The Incident Response Planner has a step-by-step playbook for deepfake fraud incidents, built around your saved plan. Part of our paid business tools.
Open the deepfake fraud playbookCommon questions
Can I still spot an AI-written phishing email?
Sometimes, but don't rely on it. AI removed most of the spelling and grammar mistakes people were taught to look for. Judge the request instead: anything that asks you to pay, change bank details, log in through a link, or keep a secret gets checked through a channel you already trust.
Is business email compromise only an email problem?
No. The same scam runs through texts, phone calls, messaging apps and video meetings, and it often moves between them. A rule that covers only email leaves the door open, so verify the request whatever channel it arrived on.
We're a small office. Would criminals bother with us?
Yes. Automation makes small targets cheap to reach, and small offices often have one person who can both set up and send a payment. That is exactly the gap a callback and a second approver close.
Will my bank refund a payment I approved?
Not automatically. If you authorized the payment, getting it back usually depends on how fast the bank can recall it or have the receiving account frozen. Call right away, file at ic3.gov, and ask your bank now what recall options your account has.
Doesn't DMARC stop this?
It helps against one trick: mail that pretends to come from your exact domain. It does nothing about lookalike domains or a real account that has been taken over, which is why a phone callback still matters.
Keep going
- ThreatHomeBusiness
Voice cloning and deepfake scams
Criminals can copy a voice from a short clip and fake faces on a video call, then use them to ask for money in a hurry. In 2025 the FBI logged 22,364 complaints that mentioned AI, with losses near $893 million, most of it in investment scams; voice-cloned 'distress' calls alone cost victims more than $5 million. The defense is low-tech: pause, hang up, and confirm through a number or channel you already trust.
Read - ThreatHome
AI-scaled romance and investment scams
Organized scam operations now use AI to translate, script and run many online 'relationships' at once, usually steering people toward fake crypto investments. Investment fraud was the costliest crime reported to the FBI in 2025 at $8.6 billion, and people 60 and older reported $3.5 billion of it. Being targeted says nothing about your intelligence; these are professional operations. The rule that holds: never send money or crypto to someone you have only met online, and talk it over with someone you trust first.
Read - GuideBusiness
Two-channel verification for payments
Money only moves after a second channel confirms it. Keep a list of known-good phone numbers, call back on those numbers before paying or changing bank details, require two people above a set amount, and know who to call in the first hour if money goes out. Among the controls finance teams use against business email compromise, that callback is the one most often rated very effective.
Read - GuideHome
The family safe word and callback rule
Pick a private word your family uses only to prove who's really calling, and agree that any urgent request for money gets a hang-up and a callback to a number you already have. The FBI recommends the secret word, and the FTC and FCC recommend the callback. Setting it up takes one family conversation.
Read - GuideBusiness
Digital hygiene for your business: the 1-50 person baseline
A small business's security baseline fits on one page: passkeys or multi-factor authentication (MFA) on every account, email and banking first; a shared password manager; automatic updates and endpoint protection; and backups you've actually restored from. Add a second-channel check before any payment change, a one-page AI rule, same-day offboarding and a short incident plan. Business email compromise alone drew $3.05 billion in reported losses to the FBI in 2025.
Read - GuideHome
Digital hygiene at home: the household baseline
A safe household runs on a few habits: one password manager, passkeys or multi-factor authentication (MFA) on the accounts that matter, automatic updates, and backups you've actually tested. AI adds one more: a family safe word and a callback rule, so a panicked call that sounds like family gets checked before anyone pays. People 60 and older reported $7.7 billion in losses to the FBI in 2025, so bring older relatives into the plan.
Read
Sources
Every figure on this page links to where it came from. We favor government, standards, and methodology-backed sources, and we take no money from anyone we cite.
- 12025 IC3 Annual Report (Internet Crime Report) · FBI Internet Crime Complaint Center (IC3) · April 2026Government · accessed October 23, 2026
- 2Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (I-120324-PSA) · FBI Internet Crime Complaint Center (IC3) · December 3, 2024Government · accessed October 23, 2026
- 3Do Spammers Dream of Electric Sheep? Characterizing the Prevalence of LLM-Generated Malicious Emails · ACM Internet Measurement Conference (IMC) 2025; Hao et al., Columbia University and University of Chicago with Barracuda Networks · October 2025Research · accessed October 23, 2026
- 4Evaluating Large Language Models' Capability to Launch Fully Automated Spear Phishing Campaigns: Validated on Human Subjects · Heiding, Lermen, Kao, Schneier and Vishwanath (arXiv:2412.00586) · November 30, 2024Research · accessed October 23, 2026
- 52026 AFP Payments Fraud and Control Survey Report: Key Highlights · Association for Financial Professionals (AFP), underwritten by Truist · April 2026Research · accessed October 23, 2026
- 6FinCEN Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions (FIN-2024-Alert004) · U.S. Department of the Treasury, Financial Crimes Enforcement Network (FinCEN) · November 13, 2024Government · accessed October 23, 2026
- 7IOCTA 2026: The evolving threat landscape. How encryption, proxies and AI are expanding cybercrime · Europol · April 28, 2026Government · accessed October 23, 2026
- 8Protecting Older Consumers 2025-2026: A Report of the Federal Trade Commission · Federal Trade Commission (FTC) · October 19, 2026Government · accessed October 23, 2026
- 9Senior US Officials Impersonated in Malicious Messaging Campaign (I-051525-PSA) · FBI Internet Crime Complaint Center (IC3) · May 15, 2025Government · accessed October 23, 2026
- 10Business Email Compromise: The $55 Billion Scam (I-091124-PSA) · FBI Internet Crime Complaint Center (IC3) · September 11, 2024Government · accessed October 23, 2026
- 11Disrupting malicious uses of AI: an update (October 2025) · OpenAI · October 2025Incident report · accessed October 23, 2026
- 12What To Do if You Were Scammed · Federal Trade Commission (FTC), Consumer Advice · June 2026Government · accessed October 23, 2026
- 13FBI Warns of Scammers Impersonating the IC3 (I-041825-PSA) · FBI Internet Crime Complaint Center (IC3) · April 18, 2025Government · accessed October 23, 2026
- 14Fictitious Law Firms Targeting Cryptocurrency Scam Victims Combine Multiple Exploitation Tactics While Offering to Recover Funds (I-081325-PSA) · FBI Internet Crime Complaint Center (IC3) · August 13, 2025Government · accessed October 23, 2026