Picture a couple three days from closing on their first house. Their phone rings, and the caller ID shows the title company. The voice is the escrow officer they met last week, friendly and a little rushed: the bank flagged the original wiring instructions, so please use the updated ones she is emailing now. The email arrives from what looks like her address, with the right file number and the right closing date. Nothing about it reads like a scam, because nothing about it is clumsy anymore.
That is the change this article is about. Closing fraud is not new, but the tools that used to give it away, the broken English, the odd phrasing, the voice that did not quite match, are being removed by the same AI that the rest of the industry is adopting. For brokerages, lenders, and everyone else at the closing table, the question is no longer whether people can spot a fake. It is whether the process works even when they cannot.
It is whether the process works even when they cannot.
The Numbers Behind the Voice
The FBI’s Internet Crime Complaint Center counted 12,368 real estate fraud complaints in 2025, with $275.1 million in reported losses, up from $173.6 million in 2024, an increase of roughly 58 percent. That figure understates the closing problem, because the FBI’s real estate category covers investment, rental, and timeshare fraud, and much diverted closing money is reported instead as business email compromise. That category alone accounted for $3.05 billion across 24,768 complaints in 2025. The same report counted 22,364 complaints that mentioned artificial intelligence, with $893 million in losses, and it describes voice cloning used to request wire payments.
Two cautions apply. These are complaints filed by victims, so they capture only the fraud that was noticed and reported, and the AI figure means a complaint mentioned AI, not that investigators confirmed it was used. Even so, the direction is plain, and it falls hardest on older owners: people aged 60 and over accounted for about $123.7 million of the real estate losses, roughly 45 percent of the total.
The Other Side of the Table
The buyer’s wire is only half the risk. In seller impersonation fraud, someone pretends to own a property, usually one with no mortgage and no one living in it, lists it for sale, and tries to collect the proceeds. The American Land Title Association’s 2026 study, released on September 14 and based on 245 title professionals surveyed in the spring, found that 59 percent of title firms had seen at least one seller impersonation attempt in the prior year, up from 28 percent in 2024. Firms reporting an attempt in the past month rose from 19 percent to 45 percent.
The targets are the properties that are easiest to fake. More than four in five firms rated vacant land at least a somewhat common target, followed by absentee-owned property, homes owned free and clear, and properties of recently deceased owners. Fifty-eight percent rated deepfake image or voice technology at least somewhat common, though the study did not measure how often deepfakes were actually used, and 87 percent said the same of spoofed contact information. One in four firms that saw an attempt also reported a paid claim. ALTA represents the title industry, and its sample is a small, self-selected one, but the doubling is hard to dismiss.
Why the Seams Are the Target
A residential closing is one of the few moments when a large sum of money moves between people who barely know each other, coordinated by four or five organizations that each run their own systems: the brokerage, the lender, the title or escrow company, sometimes an attorney, and the client’s own email and phone. Each one verifies what it can see. Nobody owns the whole chain.
Nobody owns the whole chain.
Fraud lives in the handoffs. A spoofed email from the agent to the buyer. A cloned voice from the escrow officer to the seller. A new phone number passed along in a signature line. In our judgment, this is why buying another security tool for one organization rarely closes the gap, and why the firms that do best treat verification as a shared process across the transaction rather than a feature of any one system. It is the same point the residential real estate page on our site makes about data: the weakness is not any single tool, it is the seams between them.
What Verification Has to Look Like Now
The guidance from the industry’s own bodies is consistent. The National Association of Realtors tells agents to verify wiring instructions using a phone number obtained independently, never from the email itself, and never to send wiring instructions by email. ALTA’s consumer guidance notes that it is uncommon for title companies to change wiring instructions by email and tells buyers to confirm by phone using a known number before sending funds, then to call again to confirm the money arrived.
That advice was written for a world where a phone call was proof of identity. With voice cloning, our view is that it needs three additions. First, fix the verification path at the start of the transaction: every party gets the title company’s verified number and the rule that instructions never change, in writing, before anyone is under deadline pressure. Second, verify the instructions, not the voice; a callback works only when you dial a number you already trusted, and it fails if the fraudster supplies the number. Third, put extra checks on the properties the fraudsters favor, such as vacant land, absentee owners, and properties held free and clear, with stronger identity verification for sellers you have never met.
AI belongs on the defensive side too. The same pattern-matching that makes fakes fluent can flag a changed wiring instruction, a lookalike email domain, or a listing request that fits the seller impersonation profile. If your team adopts tools like that, they touch client money and personal data, so they should pass the same gate we describe in The Best of Both Worlds, with an owner, testing, and a fallback, and our piece on handling confidential information in AI automation covers the same discipline for client communications.
When It Happens Anyway
Speed decides recovery. The FBI’s Financial Fraud Kill Chain, a process that works with banks to freeze stolen international and domestic wires, froze $679 million of $1.16 billion in attempted losses across 3,900 incidents in 2025, a 58 percent success rate, down from 66 percent the year before. The FBI’s advice is that time is of the essence: contact the sending bank immediately to request a recall, and file a complaint at ic3.gov regardless of the amount. NAR’s guidance adds that filing within 24 hours, and no more than 72, gives the best chance of recovery. ALTA publishes a rapid response plan for title companies that walks through the same steps.
Our suggestion is to treat that plan the way a hospital treats a code: write it down, put names and phone numbers next to each step, and rehearse it before you need it. The first hour is not the time to find out who calls the bank.
The first hour is not the time to find out who calls the bank.
The trust cost lasts longer than the money. A survey by CertifID, a company that sells wire verification and so has an interest in the finding, reported that 56 percent of consumers would not use the same firm again after a wire fraud incident even if all the funds were recovered, and that 22 percent of homebuyers received a fraudulent or suspicious communication during their closing. The company does not publish its method on that page, so treat the figures as indicative. The point stands without them: in a referral business, one bad closing travels.
What This Will Not Do
This is not legal or insurance advice, and it reflects public information as of October 30, 2026. The FBI figures are reported complaints, not a census of fraud, and the AI count is based on what complainants mentioned. The ALTA figures come from a small, self-selected sample, and they describe how common title professionals believe each pattern is, not measured frequency. Our recommendations are a judgment built on that evidence and on the industry’s published guidance, not a tested program. Nor is this about FinCEN’s residential real estate reporting rule, which a federal court vacated in March 2026 while the government appeals; that rule targets money laundering in cash purchases, not theft at the closing table, and its status does not change any of the risks described here.
Where to Start
Start by mapping your closing chain: every party, every channel, and every point where an instruction can change. Then write the verification rule and the response plan, and test both. If you would like an outside view, our independent assessment covers exactly that ground. Our real estate practice draws on former technology leaders from the sector, and our founder served as CIO of two residential mortgage companies. We have also written about where residential brokerage is heading in Residential Brokerage in 2040. If you would rather talk it through first, you can reach us through the contact form on our About page.
