Deepfake vs synthetic media: synthetic media is anything made or heavily changed by AI, from a generated stock photo to a cartoon voiceover. A deepfake is the part of that which shows a real, identifiable person doing or saying something they didn’t. That’s where the fraud risk sits.
Why it matters: most synthetic media is harmless, even useful. A company policy that bans “AI content” can’t be enforced. A policy against faking real people can.
| Synthetic media | Deepfake | |
|---|---|---|
| Definition | Any content generated or heavily altered by AI | Synthetic media depicting a real person |
| Examples | AI stock images, dubbed training videos, a made-up face | A fake CFO on a video call, a cloned CEO voice, a doctored politician clip |
| Is it a problem by default? | No | Usually, because it’s built to deceive |
| A case from this library | Vidoc Security Lab (2025): job candidates with made-up identities, one behind a live face filter (the company’s account) | Arup (2024): real executives faked on a video call |
Key takeaways
Deepfake scams borrow the trust people already have in someone. At Arup, Hong Kong police said every colleague on a video call except the victim was a deepfake of a real person, and the employee sent about US$25.6 million. A UK man lost about £76,000 after a Facebook ad used deepfaked video of a consumer-finance advocate and a tech entrepreneur to promote a fake bitcoin scheme. See the fake celebrity ad case.
Politics too. Hours before Maharashtra’s 2024 election, audio clips of two opposition leaders spread that forensic analysts later found were substantially AI-generated. See the Maharashtra case.
Sometimes nobody real is copied. The face, name or references are partly made up. Vidoc Security Lab says two candidates with fabricated identities reached its interviews in 2025, one recorded behind a real-time face filter. Every account of that traces back to the company’s cofounders, so treat it as their account.
Tip: when writing a company AI policy, define the harm (“never impersonate a real person without consent”) rather than the technology (“no AI video”).
In Singapore in 2026, scammers didn’t generate the officials’ faces at all. They used genuine YouTube footage of the prime minister and others and laid new speech over it. A businessman sent at least S$4.9 million. Is that a deepfake? It’s real footage, altered to show something that never happened, so most people would say yes. See the Singapore case.
No. Only AI content showing a real, identifiable person doing or saying something they didn’t counts as a deepfake. The rest is just synthetic media.
Generally no, and this isn’t legal advice. Laws mostly target specific harms, like fraud, election interference or non-consensual intimate images, rather than AI content itself.
A person who doesn’t exist, pieced together from real and invented details, often with an AI face. It’s used to open accounts or get hired. See synthetic identity.
Don’t treat any video, voice or photo as proof of who someone is. Verify requests through a channel you choose, and verify identity documents with the issuer.
Reviewed by SEE Team. Last reviewed 24 Oct 2026. How pages are researched and fact-checked. About Social Engineering Examples.
Social Engineering Examples