Safety
AI romance scams: generated profiles, voices and video
How criminals use generative AI in romance fraud, what the FBI reported in 2025, why deepfake spotting fails as a defence and the checks that still work.
Generative tools have removed the errors that used to give romance fraud away. The broken English, the recycled photograph, the refusal to appear on camera: all three are now solvable by a criminal with ordinary software. The FBI’s 2025 Internet Crime Report records 23,159 confidence and romance fraud complaints with reported losses of 929,287,469 dollars, up from 672,009,052 dollars the previous year. The defence that still works does not depend on detecting a fake. It depends on the fact that the purpose of the fraud has not changed at all: sooner or later, someone asks for money.
At a glance
Key takeaways
- Reported romance fraud losses in the United States rose to 929,287,469 dollars across 23,159 complaints in 2025, and complainants aged sixty and over accounted for 10,188 of those complaints and 584,032,745 dollars of the loss.
- Generative tools now produce plausible profile text, photographs, identity documents, cloned voices and live video, so the old quality signals such as good grammar, a clear face photo or a willingness to appear on camera no longer prove anything.
- Build a defence that does not require detecting a fake: verify through channels you choose rather than ones they supply, tell one real person what is happening, and treat any request for money or credentials as the end of the conversation.
1. What changed, in numbers
Romance fraud is not new, but its scale and polish have both moved. In the FBI’s 2025 Internet Crime Report, confidence and romance fraud produced 23,159 complaints and 929,287,469 dollars in reported losses. The two previous years recorded 17,910 complaints with 672,009,052 dollars and 17,823 complaints with 652,544,805 dollars, so both the volume and the money rose sharply.
The same report notes more than 22,000 complaints in 2025 that referenced artificial intelligence in some way, with adjusted losses above 893 million dollars across all crime types. Within romance fraud specifically, 626 complaints carried a likely AI connection and accounted for 19,041,653 dollars. Treat that last figure as a floor rather than a measurement: most victims have no way of knowing whether the person who defrauded them used a generated photograph or a real one, so an AI connection is only recorded when something made it visible.
Two other cuts of the same data are worth carrying with you. Complainants aged sixty and over filed 10,188 confidence and romance complaints with 584,032,745 dollars in losses, which is the majority of the money. And 5,925 romance complaints involved cryptocurrency, accounting for 394,787,515 dollars. Those two facts together describe the shape of the modern version: an older target, a long relationship, and an eventual investment opportunity rather than a plane ticket.
2. The four places AI shows up
The FBI has described the specific ways criminals are using generative tools in fraud, and it is worth knowing the categories rather than a list of tells.
Text is the largest one. Generated writing lets a criminal run many conversations at once, produces fluent messages in a language they do not speak, removes the grammatical errors that used to be the first warning, and fills out large numbers of fictitious profiles quickly. A chatbot can even be embedded in a fraudulent site to keep a target engaged.
Images are the second. Generated photographs make convincing profile pictures that will not appear in a reverse image search, and the same tools produce fake driving licences, staff badges and banking credentials to be sent as proof. They can also generate the specific photograph a victim asks for, including a picture holding a handwritten sign.
Audio is the third. Voice cloning needs only a short sample and is used to impersonate a relative in distress as well as to pass a phone call. Video is the fourth, and the one people underestimate: the FBI has described criminals generating video for real-time chats to prove that an online contact is a real person.
- Text: fluent messages at scale, clean translation, mass-produced profiles, embedded chatbots.
- Images: profile photographs that defeat reverse search, fabricated identity documents, photographs made to order.
- Audio: cloned voices from short samples, used for calls and for distress scripts involving relatives.
- Video: generated or manipulated faces in live calls, presented as proof of being real.
- Safety9 min
Romance scam warning signs in online dating
Recognize the relationship, identity, link and payment patterns behind romance scams—and know what to do before more money or information is exposed.
3. Why “spot the deepfake” is the wrong defence
The advice most people have absorbed is to look for artefacts: distorted hands, strange teeth or eyes, inaccurate shadows, jewellery that behaves oddly, lag between sound and mouth. The FBI does list those, and they are worth knowing. They are also a defence with a short shelf life, because each generation of tools removes another category of artefact, while the human ability to notice them does not improve.
There is a deeper problem. Detection puts you in a contest you cannot verify the result of. If you look at a face for thirty seconds and see nothing wrong, you have learned almost nothing, but you feel that you have learned something, and that feeling is the product being sold. A defence built on detection converts naturally into false confidence.
The alternative is to build your safety on the parts of the fraud that cannot be automated away. A criminal can generate a face, a voice and a document. What they cannot do is make the request for money unnecessary, because the request is the entire purpose. They also cannot easily manufacture a life that exists in places you choose to look, or survive contact with a friend of yours who has no emotional investment in the story.
4. The checks that still work
These are ordered by how much they actually tell you, and none of them require you to judge whether an image is synthetic.
Verify through channels you choose, not ones they supply. If they claim a job, find the organisation yourself and contact it through its published number. If they send a link to a verification service, do not open it; the FBI has warned specifically about verification schemes aimed at dating platform users, which harvest card details or enrol people in recurring charges. Never share a one-time code to prove you are real.
Insist that the relationship has friction. Meeting in person, at your convenience and in your city, is still the strongest signal available, and a pattern of near-misses is far more informative than any image analysis. So is talking to someone who knows them. So is a small inconsistency you introduce deliberately, such as referring to something that did not happen and seeing whether it is corrected or absorbed.
Tell one real person early. Fraud works by isolation, and the single most effective intervention is a friend who is hearing the story in real time and has no reason to want it to be true. Say at the outset who your person is, and tell them when anything about money comes up.
Finally, reduce what can be used against you. Limit the amount of your own voice and face that is public, keep social accounts private, and do not send intimate images to anyone you have not met, because generated images are also used to fabricate material for extortion.
- Contact any claimed employer, bank or organisation through a number you look up yourself.
- Never click a verification link from a match, and never pass on a one-time code.
- Agree a private word or phrase with close family, so a cloned voice in a crisis call can be tested.
- Nominate one friend who hears about the relationship from the beginning.
- Treat a first request for money, cryptocurrency or gift cards as the end of the relationship, not a topic for debate.
- Safety11 min
How to verify an online dating profile without oversharing
Use a privacy-conscious verification ladder to check whether a profile stays consistent—and understand what even a successful check cannot prove.
- Safety12 min
Online dating digital privacy: a practical protection guide
Separate account, phone, photo, location and meeting risks, then decide what to share at each stage without trying to become invisible online.
5. Live video: useful, and not proof
A video call is still worth insisting on, because it raises the cost of the fraud and because most operations will simply move on to an easier target rather than set up a live pipeline. But it has stopped being proof, and it should be described accurately when you recommend it to a friend.
Make the call harder to fake without turning it into an interrogation. Ask for an unplanned call at a time you choose rather than one scheduled days ahead. Ask them to move the camera around the room, to stand up and walk, to hold an object you name, or to write something on a piece of paper and hold it beside their face. Watch what happens at the edges of the frame, where hands meet objects, and when the light changes.
Pay attention to how the request is received rather than only to the picture. A real person may be shy, busy or self-conscious, and will still eventually do it. An operation that cannot do it will produce reasons: a broken camera, a bad connection, a remote posting, a policy at work, a family bereavement. The reasons will be sympathetic, and there will always be another one.
If a call goes ahead and something feels wrong, you do not need to prove it. Slow down, keep talking to your nominated friend, and let the money question settle the matter when it arrives.
- Safety12 min
Online dating video call safety checklist
Use video as one verification layer while protecting contact details, visible surroundings, accounts, intimate content and your next in-person plan.
6. Scripts written for religious targets
Fraud is tailored, and Orthodox Christians are a specific and easily researched audience. A generated profile can quote a psalm, name a jurisdiction, mention a parish, claim a recent pilgrimage and produce a photograph in front of an iconostasis without any of it being real. Fluent religious language is now cheap to produce, and it is doing the same work that a sympathetic biography used to do.
The patterns that recur are worth naming. A shared prayer routine established quickly, so that any doubt feels like a spiritual failure. Pressure towards marriage very early, because marriage language accelerates commitment in this community more reliably than romance language does. A claim to be Orthodox in a country where verification is hard, often combined with a story about persecution or displacement. A supposed priest, monastery or charity that needs help. And an appeal to secrecy framed as protecting the relationship from gossip.
The correction is ordinary, and it is the same one the Church would give anyway. A real Orthodox Christian has a parish, and a parish has a priest with a telephone number that you can find in a directory yourself. Ask which parish, look it up, and if the answer never resolves into a specific building with specific clergy, you have learned what you needed to know. Nobody who is genuinely who they say they are will be permanently unable to name where they receive communion.
Be equally careful with the reverse. Do not let anyone use faith to bypass your judgement, and do not use it that way yourself. Trusting God is not the same as trusting an account that appeared six weeks ago.
- Safety13 min
Spiritual manipulation in Christian dating: signs and help
Learn how religious language can be used for control, what healthy spiritual influence looks like and how to seek help without unsafe confrontation.
- Safety14 min
Red flags and green flags in Orthodox Christian dating
Evaluate repeated conduct—not religious performance or chemistry—and match your response to the seriousness of what you observe.
7. If it has already happened
Act quickly and in this order, and do not spend the first hours deciding how you feel about it. Stop all payments and contact the bank, card issuer or cryptocurrency platform immediately, because the first day matters far more than the second. Preserve the evidence before you block anyone: screenshots of the profile and conversation, the account name, phone numbers, email addresses, payment records and wallet addresses.
Then report. In the United States, file with the FBI’s Internet Crime Complaint Center, and report to the Federal Trade Commission as well. Report the account to the dating platform so it can be removed. Secure any account whose credentials were shared, and change passwords and enable two-factor authentication on anything connected to them.
Refuse recovery offers absolutely. A second approach offering to retrieve your money, whether it presents itself as a law firm, an investigator, a blockchain specialist or even as the FBI itself, is a well-documented follow-on fraud that targets people already on a list of victims. The FBI has warned about criminals impersonating the IC3 itself for exactly this purpose.
Finally, tell someone. Shame is what keeps this crime profitable, and it is what stops people reporting in time to recover anything. Victims of romance fraud include lawyers, nurses, retired police officers and priests. Being deceived by a well-resourced operation using industrial tools is not evidence of foolishness.
- Stop payments and contact your bank or platform the same day.
- Screenshot everything before blocking or deleting.
- File with ic3.gov and with the FTC, and report the profile to the dating service.
- Change passwords and enable two-factor authentication on exposed accounts.
- Ignore every recovery offer that follows, without exception.
- Safety14 min
Online dating stalking and harassment: a response plan
Prioritize immediate safety, preserve a useful record, assess technology and location exposure, and choose support and reporting options deliberately.
- Safety12 min
Online dating sextortion: prevention and an immediate response plan
A calm, victim-centered plan for stopping contact, preserving useful evidence, protecting accounts and choosing reporting and removal options.
8. Protecting the people around you
The money in this crime is concentrated among older adults, and most of them are not reading safety guides. If you have a widowed parent, an elderly relative or an older member of your parish who has recently started using a phone more, a short conversation now is worth a great deal.
Keep it warm and specific. Agree a private family word or phrase that can be used to confirm identity on a distress call, since voice cloning has made that a live risk. Say plainly that no bank, government office or relative will ever ask for gift cards or cryptocurrency. Ask them to tell you about anyone new in their life, and promise not to be scornful about it, because scorn is exactly what makes people conceal the relationship until the money is gone.
Parishes can do something too. A single announcement, or a notice in a bulletin, aimed at people who would never attend a talk about online safety, reaches an audience that nobody else is reaching. It costs one paragraph, and it is the sort of ordinary care a community is for.
Set up a defence that does not depend on detection
Do this once, before you need it, and it will hold for every conversation afterwards. None of these steps requires you to judge whether an image or a voice is synthetic, which is the point. Write the results down somewhere you will find them when you are emotionally invested, because that is when this becomes difficult.
Step 1
Write the three rules and keep them where you will see them
Write down three commitments and keep them somewhere visible: no money, cryptocurrency, gift cards or account access to anyone you have not met in person; no clicking verification links or sharing one-time codes, ever; and no intimate images to anyone you have not met. These are not judgements about any particular person and should not be revisited for a particular person. The whole design of romance fraud is to produce one exception. Decided once in advance, they cost you nothing with an honest person and end the conversation with a criminal.
Step 2
Nominate one person and tell them now
Choose a friend or relative who will hear about anyone new from the beginning, and tell them you have chosen them and why. Agree that you will mention it to them the moment money, investment, travel costs or an emergency comes up, and agree that they are allowed to say something blunt without it damaging the friendship. Isolation is the mechanism this crime runs on; a single informed outsider breaks it. If you are the person nominated, take it seriously and resist the urge to mock, because mockery pushes the story underground.
Step 3
Agree a family word for voice calls
Voice cloning needs only a short sample, and distress calls are among the most effective versions of it. Agree a private word or phrase with your parents, children and closest relatives that can be asked for on any call involving urgency or money, and make sure it is not something that appears anywhere online. Add a habit alongside it: if a call about an emergency comes in, hang up and call the person back on the number you already have. Tell older relatives about this specifically, since they are the most frequent targets.
Step 4
Practise verification on something harmless
Before it matters, run the checks once on a low-stakes conversation so they feel normal rather than accusatory. Ask for an unscheduled short video call at a time you pick. Ask which parish they attend, then look it up in a directory and see whether the answer resolves into a real building with real clergy. Contact a claimed employer through a number you find yourself. Notice how a straightforward person responds to all of this, because that is the baseline you will be comparing against later.
Step 5
Reduce the raw material available about you
Spend twenty minutes limiting what can be used to build a convincing approach or a convincing copy of you. Set social accounts to private, prune followers you do not know, and reduce the amount of your own voice and face that is publicly posted. Remove your workplace, your parish schedule and your home area from public profiles. Turn off location sharing in photographs. None of this makes you unfindable, and it is not meant to; it raises the cost of targeting you specifically, which is usually enough for an operation working at volume to move on.
Common questions
How common are AI romance scams?
The FBI recorded 23,159 confidence and romance fraud complaints in 2025 with 929,287,469 dollars in reported losses, and 626 of those complaints carried a likely artificial intelligence connection accounting for 19,041,653 dollars. The AI figure is almost certainly understated, since most victims cannot tell whether generated material was used.
Can you tell if a dating profile photo was generated by AI?
Sometimes, through artefacts such as distorted hands, odd teeth or eyes, inconsistent shadows and strange accessories, but each generation of tools removes more of these. Do not build your safety on detection; build it on verifying through channels you choose and on refusing money requests.
Does a video call prove someone is real?
No. Criminals have been documented generating video for real-time chats specifically to prove that a contact is a real person. A video call is still worth insisting on because it raises the cost of the fraud, but treat it as one signal rather than as proof.
What is a voice cloning scam?
A criminal produces a short audio clip in the voice of a relative, usually from publicly posted material, and uses it in a distress call asking for urgent money. Agree a private family word or phrase that can be requested on any such call, and hang up and call back on a number you already have.
Why do scammers send verification links?
The FBI has warned about verification schemes aimed at dating platform users, in which a supposedly free safety or identity check collects card details or enrols the target in recurring charges. Never open a verification link sent by a match, and never share a one-time code to prove you are genuine.
Are older adults really the main target?
They account for most of the money. Complainants aged sixty and over filed 10,188 confidence and romance complaints in 2025 with 584,032,745 dollars in losses. If you have an older relative who is dating online, an early and non-judgemental conversation is the most useful thing you can do.
How do scammers use religion in romance fraud?
Generated text produces fluent religious language cheaply, so a fake profile can quote scripture, name a jurisdiction and describe a parish convincingly. The reliable test is specificity: ask which parish they attend, look it up yourself in a directory, and contact it through its own published details.
What should I do if I have already sent money?
Contact your bank, card issuer or platform the same day, screenshot everything before blocking, then file with the FBI’s Internet Crime Complaint Center and the Federal Trade Commission. Secure any exposed accounts, and ignore every subsequent offer to recover your funds, as those are a documented follow-on fraud.
Sources and further reading
- FBI Internet Crime Complaint Center: 2025 Internet Crime Report
- FBI IC3: Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud
- FBI IC3: New Verification Schemes Target Users of Online Dating Platforms
- Federal Trade Commission: What To Know About Romance Scams
- Federal Bureau of Investigation: Romance scams
Pastoral practice can vary by diocese and jurisdiction. For questions about your circumstances, speak with the priest responsible for your pastoral care.
This guide provides general safety information, not legal or emergency advice. If you are in immediate danger, contact local emergency services.