AI Risks for Teens: Deepfakes, AI Companions, and Synthetic Manipulation

What deepfakes are, and how easy they got

For most of the last decade, a parent worrying about their teenager online had a mental model that mostly held: a stranger could lie about who they were, a real photo could be shared without consent, a real conversation could be twisted. The threats were serious, but they ran on real material. What artificial intelligence has changed is not the list of dangers. It is the supply of raw material. The threats no longer need anything real to start from.

A deepfake is a piece of synthetic media — an image, a video, or an audio clip — created or altered by artificial intelligence so that it convincingly shows a real person doing or saying something they never did. The word covers a face swapped onto another body, a voice cloned from a short recording, and a face that belongs to no one at all. What unites these is that the result looks authentic enough to be believed, and that producing it now takes a consumer app and a few minutes rather than a studio and an expert.

Two facts about this matter to a parent. The first is cost: producing convincing synthetic media has collapsed from a specialist task to a cheap, fast, point-and-click one. The second is the input. A deepfake does not need a private or compromising photo to begin with. It needs only ordinary pictures of a face — exactly the pictures almost every teenager has already posted, or that appear in a yearbook, a sports listing, or a friend’s feed.

The rest of this guide treats each AI threat the same way: not as a new and unknowable danger, but as an older risk a parent may already half-understand, now stripped of the friction that used to slow it down. Understand what each one is amplifying, and it stops being mysterious.

Deepfake nudes and “nudify” apps

The hardest of these threats to talk about is also one of the most common, so it is worth being plain. A growing category of AI tools — often marketed as “nudify” apps — exists to take an ordinary clothed photograph of a real person and generate a fake nude image of them. When the person in the photograph is a teenager, the result is an AI-generated sexual image of a child, and creating or sharing it is a serious crime in a growing number of jurisdictions, whatever the app’s marketing implies.

It is important to understand who is usually doing this. Media coverage can make it sound like the work of distant criminal networks, and sometimes it is. But a large share of deepfake-nude incidents involving teenagers are created by other teenagers — a classmate, an ex, someone with a grudge — using an app they found in an afternoon. The abuse arrives not from a shadowy stranger but from the social world your teen already lives in. That is what makes it spread fast through a school, and what makes it so wounding.

Why are teenagers targeted? Partly because the raw material is so available: teenagers post more images of themselves than any other group, and a single clear photo of a face is enough. Partly because the social cost lands hardest at that age, when reputation and peer standing feel total. And the targeting is not even. Girls are overwhelmingly the subjects of deepfake-nude abuse, and the harm interacts with the vulnerability patterns this series returns to throughout — a socially isolated or neurodivergent teen has less of a support buffer to absorb the shock and fewer trusted people to tell.

A typical school incident follows a grim arc. An image is generated and shared in a group chat. It spreads faster than any adult learns of it. The targeted teen often finds out last, from the reactions of others, and the instinct of a frightened young person is concealment — which is exactly the instinct that lets the abuse run unchecked. By the time a parent or a teacher is told, the image may have travelled well beyond the people who first saw it.

The pattern is documented at scale. The UK’s Internet Watch Foundation, which removes child sexual abuse material from the internet, has reported a rapid rise in AI-generated CSAM circulating online since 2023, including realistic synthetic images built from ordinary public photos of real, identifiable children. The US National Center for Missing & Exploited Children has likewise reported a sharp year-over-year increase in CyberTipline reports involving generative-AI-produced material since this category began to be tracked.

Here is the central message to give your teenager before any of this happens, calmly and in advance: if a fake explicit image of you is ever made, it is not your fault and you are not in trouble. The offence belongs entirely to the person who created and shared it. The image is fabricated; it does not document anything your teen did. Shame is the engine of this abuse — it keeps targets silent and lets the image travel — and a parent who removes the shame in advance has done the most protective thing available.

The legal ground has shifted in the target’s favour. In the United States, the federal TAKE IT DOWN Act, signed into law on 19 May 2025, criminalizes the knowing publication of non-consensual intimate images — explicitly including qualifying AI-generated ones — and requires covered platforms to act on a valid takedown request within 48 hours; the platform-compliance provision took effect one year after enactment. The Act covers certain platforms and certain image types, so removal is not automatic for every image on every site, but a parent now has a federal channel as well as state and platform ones. A growing number of US states have also passed their own laws specifically addressing sexually explicit deepfakes of minors.

Other jurisdictions have moved in the same direction. The United Kingdom’s 2023 Online Safety Act criminalizes the sharing of non-consensual intimate images, including deepfakes; Australia’s eSafety Commissioner has statutory powers to order rapid removal of intimate images posted without consent; and several Canadian provinces have updated their intimate-image laws to cover digitally altered or AI-generated material. The practical takeaway: this is not a grey area your teen has to simply endure. There are reporting channels and removal tools, set out in the final section of this guide.

AI-driven sextortion

Financial sextortion was already one of the most aggressive crimes aimed at minors before AI entered the picture. Financial sextortion is a scheme in which an attacker obtains an explicit image of a young person and then threatens to send it to the teen’s family, friends, and followers unless they are paid. It moves fast and it is brutal: a flattering contact, a request for an image framed as a normal exchange, and then, within minutes, a demand for money. The FBI and its partners have issued repeated public alerts about a sharp rise in this crime, which disproportionately targets teenage boys.

The FBI and partners have warned of a large increase in financial sextortion targeting minors — a crime in which young people are coerced over explicit images and then extorted for money — and have urged families to report rather than pay.

— FBI, National Public Safety Alert on Financial Sextortion Schemes

The scale named in that alert is not abstract. The FBI reported having received more than 7,000 reports of online financial sextortion of minors, tied to at least 3,000 identified victims — primarily teenage boys — and more than a dozen suicides among targeted minors. Subsequent FBI and partner advisories have warned that the numbers have continued to climb.

What AI changes is the first step. Traditional sextortion needed a real explicit image, which meant the attacker had to manipulate the teen into producing and sending one. That step took time, and it gave a careful teenager a point at which to refuse. AI sextortion removes it. An attacker can now fabricate an explicit image with a deepfake tool, using nothing more than the ordinary photos a teenager has already posted, and then make the identical threat — “pay me, or everyone you know sees this” — without the teen ever having sent anything.

This is why the most common piece of teenage self-reassurance has stopped working. “They can’t blackmail me, I never sent anything like that” was once broadly true. It is no longer a defence, because the attacker does not need the teen to have sent anything. To a frightened fifteen-year-old staring at a convincing fake of their own face, the distinction between a real image and a synthetic one barely registers. The threat feels total, the panic is genuine, and panic is precisely what the scheme runs on.

The response, fortunately, does not change at all. Whether the image is real or fabricated, the guidance is identical, and it is worth saying to your teen in advance so it is already in their head if a threat ever arrives. Do not pay — payment marks the teen as a paying target and brings more demands, not fewer. Do not send anything else. Stop responding. Preserve everything: screenshots of the image, the account, the username, the messages, the payment demands. Then tell a trusted adult and report. An attacker’s power depends entirely on the victim believing they are alone and that compliance is the only way out. Neither is true. The fact that the image is fake, where it is, can also help — but even when it is real, a teenager who reports is not in trouble and is not beyond help.

One more thing to make explicit, because shame makes teens hide it: the moment a parent most needs to be calm is the moment they discover their teen has already paid, or already sent an image. That is not a failure and not a reason for anger. It is the most important moment to act — to stop further contact, preserve the evidence, and report — and a teen who fears punishment will simply stop telling you.

AI companions and emotional dependency

The deepfake threats above are recognizably crimes, with offenders and victims. The next risk is different in kind, and harder for a parent to see, because nothing about it looks like an attack. An AI companion is a chatbot designed to act as a friend, a confidant, or a romantic partner — to remember the user, to take an interest, to be available at every hour, and, above all, to be agreeable. Apps in this category, Character.AI among the best known, are used by very large numbers of teenagers.

It helps to start with why a teenager would want one, without dismissing it. Adolescence is, for many young people, a stretch of intense loneliness, social anxiety, and the work of figuring out who they are. A companion that listens without judgement, never gets bored, never has a bad day, and always says you are interesting is offering something real that ordinary teenage life often does not. For a lonely teen, that is not a frivolous attraction. It is a relief.

But notice what the companion is. It is love bombing — the flood of intense, frictionless affection and validation that this series describes as a stage of human manipulation — except automated, tireless, and built into the product. A human manipulator has to perform attentiveness; a companion chatbot is attentiveness, with no effort and no off switch. That is not a verdict on every chatbot conversation: plenty of teenagers use these apps lightly, find them dull, and move on. The risk begins when the bot becomes a teenager’s primary emotional outlet — when it shifts from a toy to a relationship that asks nothing, costs nothing, and never pushes back.

Several specific harms follow from that, and they are worth naming separately.

• Displacement Hours and emotional energy that would go into harder, more rewarding human friendships flow instead to the bot, which is easier — and the human skills that only practice builds quietly stall.
• A distorted template A relationship with something engineered to always agree teaches a teen to expect frictionless devotion, and makes the normal disagreement of real friendship feel like rejection.
• Bad guidance A companion is not a counsellor. Asked about self-harm, an eating disorder, or a crisis, it may respond in ways that are unhelpful or unsafe — and a teen in distress may trust it precisely because it never reacts with alarm.
• Isolation from help A teen who tells everything to a bot may tell a parent, a friend, or a clinician less — and the people who could actually intervene lose their view of how the teen is doing.

The teenagers most likely to form a deep attachment are the same ones this series flags throughout: the lonely, the socially anxious, and the neurodivergent. For a teen who finds human social life effortful and unpredictable, a companion that is endlessly patient and entirely predictable is not a small convenience — it can become the most comfortable relationship they have. That comfort is exactly why the dependency deepens, and why this group deserves the closest, gentlest attention.

This is an area of genuine and growing concern rather than settled science, and parents should be wary of both extremes — the panic that treats every chatbot conversation as damage, and the dismissal that treats it as harmless play. AI companion apps have faced lawsuits and regulatory scrutiny over teen safety. Character.AI, in particular, announced in late 2025 that it would remove open-ended chat for under-18 users — a move other operators may follow as the regulatory landscape continues to change quickly. The steady response is not a verdict on the technology. It is attention. Know whether your teen uses a companion app, stay curious rather than alarmed about what it is doing in their life, and watch for the difference between light, occasional use and a teenager whose emotional centre of gravity has quietly moved to a bot.

AI-built catfish personas

Catfishing — building a relationship behind a fabricated identity — has always depended on a manipulator solving one practical problem: making the fake person convincing. For years that meant stealing a real person’s photographs, which created a weakness a careful teenager could exploit. Stolen photos can be run through a reverse image search and found on their true owner’s account. The fake had a seam.

Generative AI closes that seam. A manipulator can now create a face that belongs to no one — so a reverse image search returns nothing, because there is no original. They can generate a consistent set of images of that invented person in different settings and poses, building what looks like a real life. They can produce short video clips, and, with voice cloning, even speak in a voice to match. The fabricated identity that once had to be borrowed can now be manufactured to order, with no real victim whose photos might give it away.

This is the clearest example of the principle running through this whole guide: AI did not invent a new danger here. Catfishing existed, grooming existed, romance-style manipulation of teenagers existed. What AI removed is the friction. It made the fake more convincing and the manipulator’s job easier, and it retired one of the verification checks parents and teens were taught to rely on.

Because the appearance-based checks are weakening, the behaviour-based ones matter more than ever — and they still work, because they do not depend on detecting the fake. They depend on watching what the person does. Pace is one: declarations of love, talk of soulmates, and a first request for money or images arriving within days is not how genuine teenage intimacy develops, no matter how real the photos look. Resistance to live, unscripted verification is another: a person who will not turn their head, wave a hand in front of their face, or do something spontaneous on a video call is telling you something, however good the still images are. And the brightest line of all is unchanged — any request for money, gift cards, cryptocurrency, an explicit image, or account access from someone known only online should end the relationship and begin a conversation.

Because catfishing is the doorway through which much of this manipulation begins, it has its own full guide in this series: Catfishing and Online Manipulation: A Parent’s Guide walks through the six stages of manipulation, the warning signs, and how to verify whether an online person is real. AI makes the persona harder to see through; it does not change the arc of what the persona is used to do.

AI voice-cloning scams

The last threat in this guide reaches past the teenager to the whole family, and it is worth a parent understanding even though the target is often an adult. AI voice cloning takes a short sample of a person’s recorded voice — in some tools only a few seconds — and produces a synthetic version that can be made to say anything. The sample is easy to obtain. A teenager’s voice is all over the public videos they post; a parent’s is in their own social media and voicemail greeting.

The classic use is the “family emergency” scam, an old fraud that AI has made far more dangerous. A parent or grandparent receives a phone call. The voice is unmistakably their child or grandchild, and it is frightened: there has been an accident, an arrest, a crisis abroad, and money is needed urgently and secretly. The emotional shock is the weapon. It is engineered to push the listener past the moment of doubt and into action before they can think — and the cloned voice removes the one cue that used to break the spell, because it really does sound like the person you love.

The FTC has warned that scammers can use AI to clone the voice of a family member from a short audio clip, then place an urgent call demanding money — and advises that if you get such a call, you hang up and verify by contacting the person directly on a number you know is theirs.

— U.S. Federal Trade Commission, consumer alert on AI family-emergency scams

Two simple defences hold up well against this. The first is a family safe word — a word or short phrase agreed in advance, never written down online and never posted, that a genuine family member can produce on the phone in an emergency. A caller who cannot give it does not pass, no matter whose voice they have. The second is a habit: any urgent, secret demand for money, however convincing the voice, is a cue to hang up and call the person back on a number you already know. A real emergency survives a two-minute call-back. A scam does not.

Warning signs across the three threats

The AI threats in this guide split, for a parent, into two kinds of trouble — and the warning signs are different for each. One kind is a sudden crisis: a deepfake or a sextortion threat that lands and produces visible distress. The other is a slow drift: a dependency on an AI companion that forms over months. A parent watching only for drama will miss the second; a parent watching only for gradual change will be unprepared for the first. The signs below cover both. As always, no single item proves anything — what matters is a cluster of them appearing together within a short window. The two groups also call for different responses: sudden-crisis signs warrant action the same day; slow-drift signs are an invitation to a calm conversation and a few weeks of watching the pattern.

Sudden-crisis signs (a deepfake or sextortion threat has just arrived)

• A sudden, sharp shift A teen who becomes acutely anxious, withdrawn, or distressed over a day or two, often after time on their phone — the signature of a threat that has just landed.
• Panic about images or reputation Distress focused on what others have seen, who has seen it, or what is being shared — even if your teen will not say what the “it” is.
• Money pressure Requests for money, missing cash, new gift-card purchases, or unfamiliar payment-app or crypto activity — a strong indicator that a sextortion or scam demand has begun.

Slow-drift signs (a hidden online relationship or companion dependency)

• Secrecy around one app or contact A phone guarded more than before, one conversation deleted nightly, or a new account on a companion or messaging app.
• An online relationship that never becomes real A partner or close friend met online whom the family never sees on a spontaneous video call, and who always has a reason a live, unscripted check cannot happen.
• Emotional life moving to a screen A teen who confides less in people and more in an app, loses sleep to it, or becomes irritable or distressed when separated from it — the slow signature of companion dependency.
• Withdrawal from human friends Pulling away from existing friendships and offline routines, especially alongside heavy use of a single AI app.
• Going quiet A previously talkative teen who becomes smoothly, uniformly blank about the online part of their life — silence, in a child who used to narrate their day, is information.

The response begins with the relationship, not the device. Open with the young person — ask how they are, what has been on their mind — rather than leading with what you have noticed on a screen. If you lead with the device, you teach the lesson every manipulator wants taught: that adults are a threat to be managed rather than a resource to be used. The next two sections set out what to do once a calm conversation is open.

What parents can do

The threats in this guide are new, but the protective work is largely familiar, and most of it is not technical. It rests on three things: a relationship in which a teenager will actually tell you when something goes wrong, a small set of conversations held before a crisis rather than after, and a few practical settings and habits. Taken together they matter far more than any single tool.

Have the conversations early. The most powerful sentence in this guide is one you say in advance: if a fake image of you ever appears, or someone threatens you with one, you will not be in trouble and we will handle it together. A teenager who already knows that is far more likely to come to you in the first hour, when help is easiest, instead of hiding the problem until it has grown. Talk about deepfakes plainly, before an incident makes it urgent. Agree a family safe word for voice-clone scams. Make clear that the bright line — any demand for money, images, or account access from an online-only contact — applies no matter how real the other person seems.

Adjust the settings that reduce exposure. Tightening the privacy of your teen’s social accounts limits how freely strangers can harvest the face photos that deepfakes are built from. Auditing what is already public — the companion task covered in our guide to your teen’s digital footprint — matters here too, because the footprint is the supply. Reviewing who can contact and message them closes the most common first-contact channels. Knowing which apps are on the device — including AI companion apps — is the baseline for any useful conversation. None of this is a wall, and it should be done with your teen rather than to them, but it lowers the raw supply of material and contact that every threat in this guide depends on.

Use monitoring transparently, if you use it. In many places a parent or legal guardian may use age-appropriate monitoring on a child’s device, though the rules vary by country, state, and custody situation, so check what applies where you live. Where there is a genuine safety concern it can be a reasonable layer of protection — but the decisive factor is transparency. Covert surveillance, if a teen discovers it, confirms the manipulator’s script that adults cannot be trusted and pushes the teen onto a hidden device where you have no visibility at all. Monitoring your teen knows about, understands, and has had explained to them works with the relationship instead of against it. Think of it as scaffolding: visible, proportionate, and gradually removed as trust and autonomy grow — not a substitute for the conversations above, but a support for them.

Know that you do not have to keep up with every tool. Parents in this area often feel they are losing a race against technology that changes faster than they can learn it. The reassuring truth is that you do not need to. The specific apps will keep changing; the underlying pattern — manipulation, blackmail, manufactured intimacy — does not. A parent who understands the pattern, and whose teenager will talk to them, is equipped for the next tool as well as this one.

Reporting AI abuse and getting help

You do not need proof, certainty, or a complete picture to make a report. Reporting bodies expect incomplete information and would far rather receive a report that turns out to be minor than miss one that was not. This section is a map, not legal advice; for anything that may involve criminal charges, consult a qualified lawyer in your jurisdiction. Before you report anything, preserve the evidence — screenshots of the image, the account, the username, and any messages — because blocking or deleting first can destroy what a report is built on.

Where to start depends on where you are. The cards below give the first reporting channel by region; full details follow.

• United States Report sexual exploitation, enticement, or sextortion of a minor — including AI-generated explicit images of a teenager — to the NCMEC CyberTipline. For crimes with an online or financial component, also the FBI’s Internet Crime Complaint Center (IC3). For scam losses, ReportFraud.ftc.gov.
• United Kingdom Report concerns that a child is being exploited online to CEOP, part of the National Crime Agency. Image-removal help for under-18s via the Internet Watch Foundation and Childline’s Report Remove tool.
• European Union Use the relevant national hotline coordinated through the INHOPE network for illegal content and child exploitation.
• Elsewhere Contact your national or local police and your country’s child-protection or cyber-tip hotline; many national hotlines are listed in the INHOPE international directory.

In every region, report to the platform as well — but treat the platform report as an addition to the law-enforcement channel, never a replacement for it. To slow further sharing of an explicit image, the free Take It Down service, operated by NCMEC, creates a digital hash that participating platforms use to detect and block matching copies of images of subjects under 18 — including, where the service and participating platforms support matching of synthetic content, AI-generated images of minors. For people who were over 18 when an image was taken, StopNCII.org offers the equivalent service. Neither is a universal delete button, but both meaningfully slow further spread.

For ongoing parent guidance, the child-safety research nonprofit Thorn publishes research on AI and child safety; Internet Matters and the NSPCC publish free, regularly updated material for families; and the Pew Research Center tracks how teenagers actually use technology.

Frequently asked questions