AI Real Estate Fraud in 2026 — What the Data Shows About Deed Theft, Deepfakes, and Wire Diversion

The Scale, Stated Precisely

It is worth separating what is measured from what is estimated, because in fraud reporting the two get blended constantly.

The FBI’s IC3 received 1,008,597 complaints in 2025 with reported losses of $20.877 billion. Within that, the category IC3 labels real estate accounted for 12,368 complaints and $275,110,419. Business email compromise — the broader category that most closing-table wire diversion falls into — accounted for 24,768 complaints and $3,046,598,558. IC3 also flagged 22,364 complaints in 2025 that referenced artificial intelligence in some form, with $893,346,472 in associated losses.

Two cautions about those figures. First, they are reported losses, and IC3 is a voluntary reporting channel; the real totals are higher by an unknown margin. Second, the AI-referenced subtotal is based on complainants mentioning AI, which is a description of what victims noticed, not a forensic finding. It is a floor, not a measurement.

The industry-side data is more specific about direction. In ALTA’s study, the share of firms seeing an attempt in the prior month rose from 19% in 2024 to 45% in April 2026, and the share seeing three or more attempts in a single month went from 4% to 23%. That is the shape of a crime getting cheaper to attempt: not just more firms touched, but more attempts per firm.

On the consumer side, CertifID’s 2026 State of Wire Fraud report, drawn from more than 1,400 combined consumer and industry respondents and published in April 2026, found that 22% of homebuyers received a fraudulent or suspicious communication during their closing, and that 82% of consumers are already aware that criminals use AI to impersonate real estate professionals. Awareness is high. The losses continue anyway, which tells you the problem is not that people have never heard of wire fraud.

Why AI Changed the Economics, Specifically

The useful way to think about this is not “AI makes scams better.” It is that each of these schemes has one expensive bottleneck, and generative tools collapsed it.

For seller impersonation, the bottleneck was identity documents. FinCEN’s alert to financial institutions, FIN-2024-Alert004, issued November 13, 2024, describes criminals using generative AI to forge driver’s licenses and passports — altering real images or synthesizing new ones — and combining them with stolen or fabricated biographical data to build identities that survive routine verification. The same alert describes deepfake audio and video used to impersonate trusted individuals during live verification calls. Its red-flag list is instructive for how thin the tells have become: inconsistencies between a customer photo and biographical data, third-party webcam plugins appearing during a live identity check, refusal of multifactor authentication, reverse-image matches to known AI face galleries.

For wire diversion, the bottleneck was persuasion. A fraudster who has compromised or spoofed an email thread now needs the target to act quickly and not call anyone. ALTA’s respondents put spoofed contact information at 87% prevalence — the most common tactic in the study by a wide margin — and voice cloning pairs with it directly: a spoofed number that displays your escrow officer’s name, answered in something close to your escrow officer’s voice, defeats the exact verification step consumers are told to perform.

That is the part worth sitting with. The standard consumer advice for a decade has been “call to verify.” That advice is still correct, but only with a modification: call a number you obtained independently, at the start of the transaction, from a source the fraudster does not control. Calling back the number that contacted you is now a verification of nothing.

We covered the broader picture of where AI has already embedded itself in ordinary transactions — valuation, search, disclosure, underwriting — in AI is already in your real estate transaction. Fraud is the adversarial edge of that same shift, and it moves faster than the legitimate side because it has no compliance department.

Scheme One: Someone Sells a Property You Own

The mechanics are consistent enough to describe as a sequence.

The fraudster identifies a property where the real owner is unlikely to notice activity: land with no structure on it, a home owned by someone who lives elsewhere, a property with no mortgage on it, or one whose owner recently died. ALTA’s respondents ranked those targets almost exactly in that order — vacant land rated a common target by 82% of firms, absentee-owner property by 72%, free-and-clear property by 68%, and recently deceased owners’ property by 55%.

The common thread is not wealth. It is the absence of a watcher. A property with a mortgage has a servicer that sends statements and an escrow account that generates mail; a property with an occupant has someone who notices a for-sale sign. Vacant land has neither, which is why it sits at the top of the list.

The fraudster then contacts a real estate agent as the “owner,” usually from out of state, and pushes for a fast cash sale, often slightly below market, with all communication by email and text and notarization handled remotely. Identity documents are produced on request — and this is where the forged credentials do their work. If the deal reaches closing undetected, the proceeds wire out to an account the fraudster controls, and the actual owner learns about it later, sometimes much later.

The encouraging finding in ALTA’s data is that most of these attempts die before that point: 87% of the fraud that firms caught was detected during the clearance or curative process rather than at the closing table. The title search itself — the unglamorous work of confirming who actually owns the property and matching that against who is signing — remains the single most effective filter. Steve Gottheim, ALTA’s general counsel, put the industry’s posture plainly in comments on the study: “You need to have an almost all-of-the-above approach.” Firms deploying multiple detection layers, including AI tools that screen notarization video for synthetic artifacts, do better than firms relying on any single check.

When it fails, it is expensive. Twenty-five percent of firms that saw attempts also reported paid claims, and among those disclosing costs, 50% reported costs above $100,000 and another 42% between $25,000 and $100,000.

Scheme Two: The Money Goes to the Wrong Account

Wire diversion at closing is the higher-dollar, higher-frequency problem for ordinary consumers, and it has three distinct variants that fall on three different parties. CertifID’s figures below come from its own Fraud Recovery Services caseload — they describe the cases that reached a recovery firm, not a random sample of all transactions, so read them as severity rather than incidence.

Cash-to-close is the one most buyers have been warned about: a message purporting to come from the title company supplies wiring instructions, the buyer sends the down payment, and the account is not the title company’s. Payoff diversion is less discussed publicly because the victim is usually a business — a fraudulent payoff statement is substituted for the real one, and the settlement agent wires the mortgage payoff to a criminal, leaving a lien that was supposed to be released still attached to the property. Proceeds diversion hits sellers after closing, when the instructions for where to send their equity are altered.

All three share a timing signature. They land at the moment when a large, one-time transfer is expected, to a party the victim has not paid before, under deadline pressure. That is precisely the transaction profile that makes ordinary skepticism hardest to sustain.

Why Recovery Is Harder Than People Assume

The most consequential misconception about wire fraud is that a bank can simply reverse it. A wire transfer is designed to be final. What actually happens in a recovery is that the receiving institution freezes funds still sitting in the destination account, which only works if the money has not yet moved on.

The FBI’s Recovery Asset Team exists for exactly this. In 2025, RAT acted on 3,900 incidents involving $1.163 billion in attempted theft and froze $679 million — a 58% success rate. Of those, 3,574 were domestic Financial Fraud Kill Chain actions accounting for $507 million frozen, and 326 were international, accounting for $172 million. CertifID reports a comparable but not identical figure from private recovery work: 773 victims supported in 2025 and $118 million recovered, a 69% recovery rate.

Both numbers are better than most people expect, and both depend almost entirely on speed. IC3’s own instruction is unambiguous: “If you discover a fraudulent transfer, time is of the essence. Immediately, contact your financial institution and request a recall of the funds.” The practical order of operations if it happens to you is to call your bank and ask for a wire recall, then file at ic3.gov with the full transaction detail, then notify the title company and your agent — in that order, within hours, not days. Retaining a lawyer first and calling the bank second is the sequence that loses the money.

It is also worth noting what prevention looks like at scale: CertifID reports blocking 1,018 fraudulent transactions in 2025, preventing $283 million in losses. Stopping a wire before it sends is roughly an order of magnitude more reliable than chasing it afterward.

What Actually Works: A Buyer’s Checklist

These are procedural, not technical, and none of them require you to detect a deepfake.

• Establish the phone number at the beginning. At your first in-person or verified contact with the title or escrow company, write down a direct number and the name of a specific person. Use only that number for the rest of the transaction. Do not use a number from an email signature, a text, or caller ID.
• Treat all emailed wiring instructions as unverified by default. Legitimate instructions rarely change. If you receive changed instructions — especially urgent ones, especially near closing — assume fraud until you have confirmed otherwise on your pre-established number.
• Verify on a callback, not a call-in. With voice cloning and number spoofing both routine, confirming details with whoever contacted you proves nothing. You must originate the call.
• Ask your title company what it uses. Wire verification services, multifactor authentication on portals, and identity-proofing are now common. CertifID found 68% of consumers say fraud protection strongly influences which provider they choose; that is a reasonable question to ask before you engage one.
• Send a small test wire when the option exists. Not all agents accommodate it, but confirming receipt of a nominal amount before sending six figures costs a day and eliminates the most expensive failure mode.
• Know the first two calls. Bank, then IC3. Decide that in advance, because the window is measured in hours.

A separate warning for anyone searching for housing rather than buying: the FTC’s December 22, 2025 data spotlight counted nearly 65,000 rental scam reports and $65 million in reported losses from January 2020 through June 2025, with a median individual loss of $1,000. Roughly half of the reported scams originated from fake ads on Facebook and about 16% from Craigslist, and adults aged 18 to 29 accounted for 46% of reports with losses — roughly three times the rate of other adults. The FTC did not attribute these to AI-generated content, and it would be wrong to claim it did; the relevant point is simply that listing-stage fraud is heavily concentrated on social platforms and on younger renters. Never send a deposit for a property you have not seen in person, and never to someone who cannot let you in.

What Actually Works: A Homeowner’s Checklist

If you own property you do not live in — land, an inherited house, a rental, a second home — you are in the demographic ALTA’s respondents identified as the primary target, and the defenses are different.

• Enroll in your county recorder’s property fraud alert. Most counties now offer a free notification when a document is recorded against your name or parcel. It is the closest thing to a smoke detector for deed fraud, and the enrollment takes minutes. Search your county recorder or clerk’s office directly rather than through an ad.
• Do not confuse that with a paid “title lock” product. These services generally monitor and notify; they do not prevent a fraudulent recording, and the free county alert covers the same detection function. Read carefully what a paid subscription actually promises to do.
• Check your parcel’s recorded documents periodically. Most county records are searchable online at no cost. Once or twice a year is enough to catch something that a mailed notice missed.
• Keep the tax mailing address current. Notification programs mail to the address on the tax roll. An out-of-date address defeats the entire mechanism.
• Maintain an owner’s title insurance policy and know what it covers. An owner’s policy is the financial backstop when a forged conveyance clouds title; the coverage question in a forgery claim is fact-specific, so read the policy and ask your insurer directly rather than assuming.
• Treat unsolicited offers on vacant land with suspicion in both directions. If you receive an unsolicited cash offer, verify who is behind it. And if you are a buyer being offered vacant land cheap by a remote seller in a hurry, you are looking at the profile of the most common fraud in the study.

If you are weighing whether to hold a property you have moved out of, the monitoring obligation is a real, if small, line item in that decision — one of several we walk through in the rent-it-out-or-sell-it framework.

Where the Rules Are Going

Regulation is moving, unevenly, in two directions: detection duties on institutions, and notification rights for owners.

On the institution side, FinCEN’s November 2024 alert did not create a new rule so much as tell banks what to look for and how to flag it — suspicious activity reports involving deepfake media are to carry the keyword “FIN-2024-DEEPFAKEFRAUD.” That produces a data trail regulators did not previously have, which is usually the precursor to more prescriptive requirements.

On the owner side, California enacted SB 255, which requires every county to establish a recorder notification program by January 1, 2027. Under it, county recorders or their designees must mail notice within 30 days of recordation to the parties executing deeds, quitclaim deeds, mortgages, and deeds of trust, at the property tax mailing address. Counties may also offer electronic notification. The statute shields recorders from liability for failing to give notice, which is a meaningful limit on how much protection it actually confers.

The honest assessment of a 30-day mailed notice is that it is a detection mechanism, not a prevention mechanism. It shortens the interval between a fraudulent recording and the owner finding out, which matters enormously for unwinding the transaction — but the money has usually already moved. It is a floor under the problem, not a solution to it. Other states are moving in similar directions; if you own property, check what your own county already offers, because in many places a free alert program predates any statutory mandate.

The Honest Read

None of this is a reason to avoid transacting. The market itself is unremarkable at the moment: the 30-year fixed averaged 6.76% for the week of September 10, 2026, up from 6.71% the prior week and 6.35% a year earlier (Freddie Mac), and existing-home sales ran at a 3.98 million annualized rate in August 2026, down 1.2% from a year earlier, with a median price of $429,100 and 4.9 months of supply — the highest inventory reading in over a decade (National Association of Realtors, released September 10, 2026). Roughly four million households a year complete these transactions without incident.

What has changed is where the burden of verification sits. For most of the modern era, the forgery bottleneck did quiet, unpaid work on your behalf: bad fake IDs were bad, and cloned voices did not exist. That layer is gone, and nothing has replaced it automatically. The layer that replaced it is procedural — a phone number you wrote down at the start, a callback you originate, a county alert you enrolled in, a title search someone actually performed.

The reassuring part of the 2026 data is that this works. Eighty-seven percent of caught fraud was caught in the title clearance process, before closing. The overwhelming majority of attempts fail. The failures that matter are concentrated in a narrow set of moments: a changed wiring instruction accepted without an independent callback, and an unwatched property whose owner was not notified. Both are addressable this week, at no cost, by anyone reading this.

The one thing not to take from the numbers is complacency about detection. The premise underneath every consumer-facing warning about deepfakes — that a careful person can tell — is deteriorating, and the defenses that survive are the ones that do not require you to tell.