--- id: scams-and-fakes title: AI scams and deepfakes level: 2 modes: [day-pack, rucksack] moment: "The person gets an urgent call, voicemail, video or message that sounds like someone they know or an organisation and asks for money or a code, or they ask whether a voice, photo or message is real." keywords: [scam, deepfake, voice clone, fake video, phishing, ring back, code word, urgent request, impersonation, fake image] source: "research/2026-10-20-concepts/scams-and-fakes.md; research/2026-10-23-essentials/batch-5-facts.md; mapping (product/learning-system/MAPPING-1-2.md): FAQ 2.12 (home), 4.13, 4.19; tips T082, T083; strand P" --- ## What it is AI can copy a voice from a few seconds of recording, fake a face on video and write a perfectly spelled message, so old tells such as bad spelling or a strange voice no longer work. Judge a request by the rush to act, especially if it wants money or a code, and check it through a number you found yourself. If a voicemail in your daughter's voice asks for $3,000 now, hang up and ring her on the number you already have. ## Tips - Hang up and ring back on a number you found yourself when a call rushes you to pay, because voices and caller IDs can both be faked. - Judge a message by the rush to act, not by spelling or how it looks, because AI now lets scammers write flawless messages and copy voices and faces. - Be wary of any urgent request to pay by bank transfer, cryptocurrency or gift card, because those are the payment routes scammers prefer. ## Questions people ask - Q: How do I tell what's real, and spot AI scams, fake voices and fake images? A: Not reliably by looking or listening, and old tells such as bad spelling no longer hold. Judge by the rush to act, check through a number you found yourself, and do not trust an AI app's word that something is real, because AI apps get fooled too. - Q: Can I tell a cloned voice by ear? A: Not reliably, since in a lab study listeners judged a clone to be the same person about 80 percent of the time. Ring back on a number you already have. - Q: Is a family code word enough? A: No, it is one layer. Pick a phrase nobody could guess, agree it in person, never text it, and still ring back. - Q: What if it really is my son in trouble? A: It could be, so hang up and call him back on a number you already have for him, or call someone close to him, and pay nothing and share no code until you have spoken to him or someone close to him. Do not ring a number the caller gave you. - Q: I already paid. What now? A: In Australia, call your bank first, then IDCARE, and report it to Scamwatch. ## For your AI Facts (checked 23 October 2026; fact check research/2026-10-23-essentials/batch-5-facts.md; research note research/2026-10-20-concepts/scams-and-fakes.md). Australian contacts; elsewhere differs. - Voices: in a lab study, listeners judged a cloned voice to be the same person about 80 percent of the time (Scientific Reports, 31 March 2025, https://www.nature.com/articles/s41598-025-94170-3); longer, unscripted speech helped. Scamwatch says a few seconds of recording is enough, and to call back on the person's real number or an organisation's number from its official website or app (https://www.scamwatch.gov.au/stop-check-protect/help-to-spot-and-avoid-scams/how-scammers-use-technology-and-ai). - Spelling is no longer a tell: AI-written phishing was clicked as often as experts' (54 percent, 5 June 2026, https://www.schneier.com/academic/archives/2026/06/evaluating-large-language-models-ability-to-automate-spear-phishing.html), though some official guides still list poor spelling. - Payments: in 2025 Australians lost most through cryptocurrency and bank transfer, the method scammers ask for most; gift cards smaller (National Anti-Scam Centre, https://www.nasc.gov.au/system/files/targeting-scams-report-2025.pdf). - Code words: the FBI (https://www.ic3.gov/PSA/2024/PSA241203) and Westpac (2 April 2026) suggest a family phrase to confirm identity; the UK's Starling Bank says never share it digitally. It is one check, not proof. - AI apps judge fakes poorly: ChatGPT missed 92.5 percent of unmarked AI videos in one test (NewsGuard, 22 January 2026, https://www.newsguardtech.com/special-reports/top-ai-chatbots-dont-recognize-ai-generated-videos/), and chatbots called AI images real (Brennan Center, 5 October 2026, https://www.brennancenter.org/our-work/analysis-opinion/chatbots-still-bad-spotting-ai-generated-images-despite-new-law). - Already paid (Australia): call the bank on the number on the card, app or website, then IDCARE (1800 595 160, weekdays 7am to 7pm Australian Eastern time, https://www.idcare.org/), and report at scamwatch.gov.au (https://www.scamwatch.gov.au/protect-yourself). Teach it like this: when a call or message rushes them to pay or share a code, say plainly to stop and check through a number they found themselves, never the one given. Never give or look up a number for them: it must be one they already have, or the one on the organisation's official website or app. For a voicemail, do not reply to it or ring the number it came from. Never say you can tell them whether a voice, video or image is real.