IntegrAuth Academy
Identity, security and practical AI — free, from zero. Byte-sized lessons, real diagrams, hands-on labs, zero fluff. Two programs, each with its own exam and certificate.
English is hard? Your browser can translate this page into Tamil, Telugu, Kannada, Malayalam or Hindi. Tech words stay in English so you learn the real names.
How?
- Chrome on phone: ⋮ menu → Translate
- Chrome on computer: right-click → Translate to …
- Edge: the translate icon in the address bar
Deepfakes and AI scams
Priya's phone shows her manager's number. The voice sounds like her manager. The caller wants a secret payment before five. Your goal: check a request another way.
🧪 Interactive lab — enable JavaScript to play with this one.
- Urgent, secret, odd payment? Hang up. Call back on a number you already have.
- Get AI apps only from the real maker. Real AI can't promise you profits.
- A label you find shows where a file came from. No label proves little.
- A detector score is a lead, never proof.
What AI can fake
AI can make a fake video, sound clip or image of a real person. This is a deepfake. An AI copy of one person's voice is a voice clone. In 2023, a US consumer agency warned that a short clip may be enough. AI also writes smooth messages. So bad grammar no longer gives a scam away.
Check another way
See also Images, voice and video. Most scams push 3 things. Urgency: "before five." Secrecy: "don't tell anyone." An odd payment or request: new bank details, gift cards, a code. See 2 of these together? Slow down.
In India, a common one is the "digital arrest" call. A caller in police uniform says you are under arrest and must pay. Police do not arrest anyone by video call. Hang up and call 1930.
Check through a route the scammer does not control. This is out-of-band verification. Hang up and call a number you already have. Never use a number or link from the message.
Labels on files
Some files carry a signed record of what made them and how they were edited. This is Content Credentials (C2PA). Some AI tools hide an invisible signal in what they make. This is an AI watermark. A missing label proves almost nothing.
Now follow some files. Guess what each check shows.
🧪 Interactive lab — enable JavaScript to play with this one.
Go deeper: scam patterns and a real case
Two work scams are common. A fake "boss" on a video call asks for urgent, secret transfers. A fake job candidate uses a face swap to get hired and get into company systems.
The best-known case is the engineering firm Arup. In early 2024, a worker in its Hong Kong office joined a video call. The finance chief and several colleagues looked and sounded right. Hong Kong police say they were all fake. The worker made 15 transfers, about US$25.6 million.
Go deeper: fake AI apps and money promises
Scammers also sell fake AI:
- Copycat AI apps. They copy a famous AI name and logo. Some charge a high weekly fee. Some steal data. Check the developer's name, or use the maker's own website.
- AI add-ons. A browser extension (a small add-on for your browser) may ask to read every site you visit. Some have copied people's AI chats. Install few.
- "AI trading bots." They promise sure profits. No real AI can. US regulators warned about them in 2024.
- Paid "AI income" courses. They sell easy money, sometimes as packs of "secret prompts." Buyers often lose money. In 2024, a US consumer agency acted against several such schemes.
Lost money? Report it fast. In India, call 1930 or use cybercrime.gov.in. In the US, use reportfraud.ftc.gov.
Go deeper: more habits
- Agree a family code word. The FBI recommends a secret word or phrase to prove who you are. Keep it offline.
- Let the process protect people. At work, two people approve large payments. New bank details are checked by calling a known contact.
- A voice is not a password. Avoid sign-in by voice alone (identity verification).
Moving a meeting needs no check. A request for money, data, codes or access always gets one.
Go deeper: why eyes and detectors fail
In a 2022 study, people judged AI-made and real faces. They were right 48.2% of the time, about a coin toss. In a 2023 study, listeners missed about 1 speech deepfake in 4.
A good fake may show no odd hands or out-of-sync lips. Seeing no glitch proves nothing. A deepfake detector gives a score, not a verdict.
Detectors for AI-written text do worse. A 2023 Stanford study tested 7 detectors on essays by people writing English as a second language. On average, 61% were wrongly marked as AI. So a score is a reason to look closer. It never proves that a person cheated.
Go deeper: what labels prove
| Signal | Found? It tells you | It can't tell you |
|---|---|---|
| Content Credentials | Who signed the record. That the file has not changed since. | If the scene or caption is true. Many social sites and screenshots drop the record. |
| AI watermark | A tool that adds that watermark made or edited it. | Anything about tools that add no watermark. |
| Detector score | The content looks like what the detector learned. | Proof of anything. |
One known watermark is SynthID from Google DeepMind. As of October 2026, the Gemini app can check images and videos for it. It only tells you if Google AI made or edited the file. Since February 2026, India's rules ask online platforms to label AI-made content. The EU's rules are in AI rules at work.
Don't doubt everything. Someone caught on a real recording can now say "that's a deepfake." This is the liar's dividend. Never accuse a person on a detector score alone.