Scams, Deepfakes & Fake Content

Fakes just got cheap

For most of history, a convincing fake was expensive: a forged photo needed a skilled retoucher, an impersonated voice needed a talented mimic, a polished scam letter needed a fluent writer. AI collapsed all of those costs at once:

• Voices can now be cloned from a few seconds of audio — a clip from a social media video or even a voicemail greeting is enough to make a voice say anything.
• Images that look like photographs can be generated from a text description in seconds, for free.
• Video of a real person appearing to say things they never said — a deepfake — has gone from research labs to phone apps.
• Scam messages can be written by AI: fluent, personalised to you, error-free, in Bangla or English, and produced by the thousand.

Here's the mental shift this page asks of you, and it's a single sentence: "it looks real" and "it sounds like them" are no longer evidence. That used to be a reasonable test. It isn't anymore. What still works is checking where a thing came from — and that skill is the whole page.

The scams making the rounds

Four patterns account for most of the AI-powered fraud happening right now:

• The voice-clone emergency call. Your phone rings. It's your son's voice, or your sister's — upset, in trouble, in an accident, arrested, robbed while abroad. Money is needed right now, and please don't tell anyone. The voice is cloned; the emergency is fiction. Elderly parents are the favourite target, and the emotional punch of a loved one's voice is exactly what the scam is built on.
• The celebrity investment deepfake. A video circulates: a famous businessman, a cricketer, a TV anchor enthusiastically endorsing a crypto platform or investment scheme that "doubles your money". The face is real, the voice is close, the endorsement never happened. These run as paid ads on legitimate social platforms, which makes them feel vetted. They aren't.
• The too-perfect stranger. Romance profiles with flawless, AI-generated photos and attentive, endlessly patient chat (also AI). Job offers with great pay for vague work, where the "interview" goes suspiciously smoothly and then a "registration fee" or your ID documents are requested. In both, AI plays the charming human until the ask arrives.
• The polished phishing message. For decades we taught people to spot scam emails by their broken grammar. That tell is dead. AI-written phishing — fake messages pretending to be your bank, courier service, or office IT — is now grammatically perfect, correctly formatted, and can reference your real name and workplace. Judge messages by what they ask you to do, never by how well they're written.

Your defences

The good news: none of the defences require technical skill, and all of them are free.

• Verify through a second channel. This is the master move. Voice call claiming to be your nephew? Hang up and call the number you have saved for him. Email from "your bank"? Don't tap the link — open the bank's app or call the number on your card. WhatsApp message from "your boss" asking for gift cards? Walk to their office or call them. The scammer controls one channel; they almost never control two.
• Agree a family code word. Pick a word or question only your family knows — an old pet's name, what you cooked at some family event. Any emergency money request, in any voice, must answer it. Set this up this week; it takes five minutes at dinner and it beats every voice-cloning tool on earth.
• Learn the triple red flag: urgency + money + secrecy. Each alone can be innocent. All three together — "act immediately", "send money / share a code", "don't tell anyone" — is the signature of nearly every scam ever run, AI or not. Treat the combination as an alarm, not a coincidence.
• Reverse image search. For a suspicious photo — a profile picture, a "news" image — search by image (Google Lens and similar tools do this from your phone). If the "Canadian engineer" who's courting your aunt has a profile photo that appears on stock-photo sites, case closed.
• Check official sources for "news". A shocking video of a politician, a "leaked announcement", a miracle cure: before believing or forwarding, look for it on two or three established news outlets. Real events get covered everywhere within hours. A story that exists only as a forwarded video exists only as a forwarded video.

Spotting AI images — and why source beats pixels

You'll see checklists for spotting AI-generated images, and they're worth knowing: hands with odd fingers, garbled text on signs and clothing, mismatched earrings, lighting that doesn't agree between face and background, skin smoothed to plastic perfection, backgrounds that melt into nonsense when you zoom.

Know them — and then hear the honest truth: detection is losing the race. Each of those tells was reliable for one generation of tools and gone the next. Today's best fakes have convincing hands. Building your safety on "I can spot a fake" means rebuilding it every six months, and one day being wrong.

The durable skill is different: verify the source, not the pixels. Who posted this first? Is it on any credible outlet? Does the person or organisation it shows confirm it on their official account? A real photo has a traceable origin; a fake has a chain of forwards that starts nowhere. Provenance ages well. Pixel-squinting doesn't.

Fakes in your feed

Not all AI fakery wants your money. A growing share of what circulates on social media — outrage-bait images, fake quotes attributed to famous people, "historical photos" that never happened, entire AI-written news sites — wants your clicks and shares. Each share you give a fake lends it your credibility; being the aunt or uncle who forwards fakes costs you exactly that.

The habit that fixes it is small: before sharing, check twice. Thirty seconds — who posted this originally, and does any real outlet carry it? If you can't confirm it, don't forward it; if you already did, delete and say so. And interestingly, your own AI chatbot is a decent ally here — paste in a suspicious message and ask for an analysis:

I received this message and I'm not sure it's genuine. Analyze it like a scam expert:
1. List every red flag you see (urgency, money, secrecy, strange links, requests for codes or IDs).
2. Tell me what this scam type is called, if it matches a known pattern.
3. Suggest the safest way for me to verify it through official channels.

Here's the message:
[paste the suspicious message here — remove any personal details first]

One caution to complete the picture: the AI's analysis is an aid, not a verdict — it can miss things, as When AI Gets It Wrong explained. And note the bracket line: strip out account numbers or personal details before pasting, exactly as Privacy — What Not to Share taught. For checking suspicious "news" claims with web-search mode, the techniques in Research with AI apply directly.

Key takeaways

• AI made fakes cheap: cloned voices from seconds of audio, photoreal images, deepfake videos, and flawless phishing. "Looks real" is no longer evidence.
• The old "bad grammar" test for scam messages is dead. Judge messages by what they ask you to do, not how well they're written.
• Master defence: verify through a second channel — hang up and call back on the number you have saved. Set a family code word this week.
• Urgency + money + secrecy together = assume scam. Never share an OTP with anyone, ever.
• Spotting fake pixels is a losing race; verifying the source — original poster, official accounts, established news outlets — is the skill that lasts.
• Before forwarding anything shocking, check twice. This is not paranoia — it's the same street-smarts you've always had, on a new street.