AI Romance Scams: Criminals Now Fake Love at Scale

Romance scams stopped being a lonely heartstring crime and became an industrial one. The FBI’s Internet Crime Complaint Center counted $929,287,469 in reported confidence and romance fraud losses in 2025, and the FTC found that nearly 60% of people who lost money to a romance scam said it started on a social media platform (FTC, 2026; IC3, 2026). Generative AI now does the hard part: it writes fluent messages, produces believable photos, and can even power a live video call, so a single operator can court a dozen targets at once instead of one.

Key takeaways

• FBI IC3 counted $929 million in romance fraud losses in 2025, its third straight year of growth.
• 60% of romance scam losses involved contact that started on social media.
• Generative AI removes the spelling errors and photo tells victims used to rely on.
• Live video calls can now be faked, so "prove it on camera" is no longer a reliable test.
• The fastest recovery move is reporting to your bank, the platform, and IC3 within the first 48 hours.

How the numbers got so big

Reported losses compound for a grim reason: the work scales. Manual romance scammers had to type every message and send each photo by hand, which capped how many victims one person could manage. AI inverts that economy. A scripted assistant can hold parallel conversations in fluent English, translate mid-sentence, and regenerate a "selfie" to match a new persona in seconds, so the criminal’s marginal cost of a second victim drops toward zero (FBI Boston, 2026).

The FTC’s contact-method data shows where the traffic lives. Of everyone who reported losing money to a scam that started on social media in 2025, romance schemes were among the top categories, and nearly 60% of romance scam victims said the platform was where contact began (FTC, 2026). Social media outcomes like investment romance, crypto referral, and emergency wire all funnel into the same app the first message was sent from.

What changed when AI got into the script

The classic scammer tells used to give themselves away. Typo-crammed messages, stocky photos, video call refusals, and "my camera is broken" were the tripwires every safety guide taught. AI deletes the tripwires. Language tools flatten the grammar into native-sounding prose, and image generators produce face after face with no reverse-image hit, so the old "run their photo through Google" test comes back clean (FBI Jacksonville, 2026).

Police bulletins now warn that criminals use generative AI for translations to hide foreign origins and generate photos to prove "they" are real. That closes the two checks families used to run, which means the new detection has to live in behavior, not in files. Urgency, secrecy, and money requests still follow the same old rhythm, and that rhythm is what trained observers watch for.

The video call that breaks every old rule

Face-to-face video was the last sacred test, and it is gone. Real-time deepfake tools can now map a generated or stolen face onto a live webcam feed, run lip-sync on the fly, and keep a conversation going through a voice filter. FBI field offices have documented victims who "saw" the person on camera and still lost savings to a fabricated face (FBI Boston, 2026).

The practical markers shift to the machine: ask the person to turn their head slowly side to side, wave a hand in front of the lens, or say your name mid-sentence, then pause a frame. Render artifacts surface as glitchy ears, glassy skin, and mouths that slip out of sync for half a beat. None of these catches everything, but they raise the production cost, and criminals drop targets who make the show expensive.

Why older adults carry the heaviest losses

Victims aged 60 and older lost $584 million to confidence and romance fraud in 2025, roughly 63% of the entire category, while representing a smaller share of complainants (IC3 Elder Fraud section). Social isolation plus unfamiliarity with deepfake tools line up with the highest-value targets. These are the relationships children find out about only after the money is gone.

The crypto pivot that multiplies the damage

The modern script almost always bends toward cryptocurrency. The "lover" mentions a trading platform, sends a link, and lets the victim win a few small trades before the withdrawal fee appears. Crypto transfers are fast and near-irreversible, and they exit through wallets exchanges cannot claw back, which is why IC3 tracks billions lost to crypto-linked schemes each year (IC3, 2026).

The romance is the honeypot and investing the hook. By the time the victim tries to withdraw, the platform demands "tax," "collateral," or a minimum balance, and the paramour is suddenly unreachable. Families who find a relative deep in one of these loops should treat every future transfer request as engineered rather than accidental.

Five red flags AI cannot hide

• A relationship that escalates to declarations of love within days and rushes off-app.
• Refusal to meet in person on any schedule, with reasons that keep renewing.
• Requests for money framed as emergencies, customs fees, or investment "minimums".
• Pressure to keep the relationship secret from family and friends.
• A sudden invitation to a crypto or trading platform with no paperwork.

These five markers survive no matter how polished the AI gets, because they are behavioral choices, not production qualities. A fluent grammar and a generated face change nothing about the fact that a stranger is asking for urgent, secret, irreversible money.

What the platforms say they are doing

Social media companies publish anti-romance-scam pages, remove profiles on report, and cooperate with law enforcement, but the reporting burden sits with the user, and the under-reporting rate is high because victims feel embarrassed (FBI Boston, 2026). Dating apps have added AI-detection for verified photos, yet verification only proves a still is genuine, not that the human behind it is honest.

Realists treat platform trust and safety as the backstop, not the primary defense. The primary defense remains a second pair of eyes: a family member or friend who sees the screenshots, knows the red flags, and has permission to ask the hard questions before any money moves.

What to do if a relative is already entangled

Start from care, not accusation. Victims of romance fraud are often convinced the relationship is real, so "you are being scammed" instantly makes them defensive. Instead, say you are worried this person has done this to others, show the IC3 and FTC pages, and offer to test the "investment platform" by asking for a real withdrawal of a small amount.

If money has already moved, act in the first 48 hours. Report to the bank or credit union, the platform where contact started, the FTC at reportfraud.ftc.gov, and IC3 at ic3.gov. Each report lands in a different database, and every one improves the chance of a freeze, a recovery, or a future warning to the next victim.

The damage that never appears in the spreadsheets

The dollar figures understate the cost because they only count wire transfers and credit lines. Victims describe a second loss that does not make the IC3 table: the shattering of trust in their own judgment, the shame that keeps them silent, and the isolation that grows as they pull away from friends who "would not understand." FBI field offices note the embarrassment factor explicitly as the main reason fraud against the elderly goes unreported (FBI Boston, 2026).

The emotional aftermath changes how the family should respond. A parent who lost $100,000 to a fabricated relationship does not need a lecture on scam hygiene, they need a plan that separates the fraudster from the money trail and reinforces that the victim is not stupid. Support groups and financial-counseling lines exist specifically because recovery is as much about dignity as about balances.

How one report can stop twenty future victims

The reporting loop is the under-appreciated defense. When a victim files with IC3, the complaint joins a dataset that the FBI and platform safety teams mine for coordinated clusters, shared phone numbers, and the same bank-account routing patterns across dozens of cases. A single detailed narrative can link a criminal network that has been working a region for months (IC3, 2026).

The burn-the-link habit works the same way at family scale. Victims should keep one screen capture of every platform profile, one screenshot of the request messages, and the exact wallet or transfer reference. That package turns a hopeless "I got scammed" into an investigation that can claw money back or warn the next target before their first transfer.

And the cheapest intervention remains the conversation that happens before the money does. A household rule that "anyone you have met only online, before anything is ever sent, gets one call to a trusted third person" has stopped more romance fraud than any filter, detector, or country regulation to date. AI made the scammers fluent, and it cannot make them honest.

Frequently asked questions

Related coverage

• How to verify AI answers before you trust them
• Why AI chatbots hallucinate and how to catch them
• AI slop is eating the web
• A field guide to digital minimalism
• FTC Data Spotlight: social media scam losses in 2025
• FBI IC3 2025 Annual Report PDF
• AP: romance scams hit a record high in the US