Voice and video are no longer sufficient proof
A familiar face, voice or speaking style can be copied from public material and placed inside a false call, message or recording.
Technology, platforms and channels
AI, deepfakes and synthetic impersonation
AI deepfakes can create or alter convincing voices, images and video, making impersonation faster, cheaper and easier to personalise. Appearance and sound alone cannot confirm identity. Verify the person, event and requested action through a separate route you already trust.
Page T-005Guidance pageReviewed 29 August 2026
Key Guidance
Do not try to decide authenticity from appearance or sound alone. Break the contact and verify the person, event and requested action through a separate route you already trust.
Start with Key Guidance for the short answer. Then read the sections that match what happened and what you need to do.
Understand the issue
Why this matters to you
Synthetic media changes the strength and cost of impersonation. It can make fabricated evidence easier to produce and more personally convincing, while also making genuine media easier to dismiss as fake.
Personalisation can scale
AI can adapt language, tone and context quickly, allowing the same campaign to feel individually written for many recipients.
Detection tools have limits
Synthetic-media detectors can be inconsistent and may not give a reliable answer for every format, edit or low-quality sample.
The request still reveals the process
Pressure, secrecy, changed payment instructions, supplied verification routes and refusal of independent contact remain meaningful even when the media looks convincing.
Consequences and movement
What may happen next
AI can support several different forms of deception. Separate the synthetic artefact from the wider situation so one impressive element does not dominate the decision.
Family or emergency impersonation
A cloned voice may claim distress and demand secrecy, urgency or payment before normal family checking can occur.
Workplace authority
Synthetic audio or video may reinforce a request that appears to come from an executive, supplier or colleague.
Investment and celebrity endorsement
Altered video and fabricated interviews may lend authority to a false opportunity or platform.
Identity and relationship building
Generated photos, video, documents and conversation can maintain a synthetic persona over time.
Practical response
What you should do
The safest checks do not depend on spotting visual glitches. They test the claimed relationship and requested action outside the suspicious channel.
- Pause the requested actionDo not transfer money, share a code, disclose information or change a process while the contact controls the timing.
- Use a known routeCall the person on a saved number, reach the organisation through its official directory or ask an authorised colleague in a separate channel.
- Test the situation, not only the mediaAsk whether the request fits normal authority, payment, approval and confidentiality procedures.
- Use agreed verificationFamilies and teams can confirm a safe word, call-back rule or second-person approval for high-risk requests.
- Preserve the original materialKeep the message, recording, account details, timestamps and payment instructions. Avoid repeatedly compressing or editing the evidence.
- Warn connected peopleIf your voice, image or account is being used, tell likely contacts through a trusted route and report the impersonating account or content.
Look more closely
A convincing voice or video still needs independent verification
Synthetic media can make a person appear present without proving who created the content or why. Verify the person, event and requested action through a separate route before relying on appearance or sound.
A call, message, advert, meeting or relationship creates access.
Urgency, affection, authority, opportunity or fear makes the contact matter.
Synthetic media makes a person, event or endorsement appear present and credible.
The contact controls how identity is checked and where action takes place.
Payment, sign-in details, confidential data, approval or account access becomes the ask.
Selected history
Historical cases and lasting lessons
Synthetic executive video conference
Hong Kong authorities described a case in which a worker received a phishing email, joined a fabricated pre-recorded video conference appearing to involve a chief financial officer and authorised transfers of about HK$200 million.
Lasting lesson: The decisive failure was not simply believing the video. The payment request, confidential framing and absence of an independent approval route allowed the Mask to control the Path.
Deepfake investment and identity campaigns
Authorities reported synthetic celebrity or official endorsements, face-swapped identity documents and AI-assisted impersonation across several fraud types.
Lasting lesson: Deepfake is a technique used inside different scams, not a single scam category with one warning sign.
Transferable understanding
Patterns over time
Realism reduces friction
A familiar voice or face can suppress the pause that an unusual text-only request might otherwise create.
Channel control matters more than perfect media
A criminal may move from synthetic video to ordinary messaging once authority has been confirmed.
AI can support both mass and targeted scams
Automation scales broad campaigns, while public media and breached data can personalise selected targets.
Authenticity is not the same as legitimacy
Even genuine media can be taken out of context, and a genuine person can still make a compromised or mistaken request.
Plain-language answers
Common questions
Can I spot a deepfake by looking for strange eyes or lips?
Sometimes artefacts are visible, but quality varies and detection by appearance is unreliable. Use independent contact and process checks as the primary defence.
Will a deepfake detector give me a definite answer?
Not reliably in every case. Detection tools may help investigation, but a public decision should not depend on one automated score.
What if the person answers a personal question?
The answer may be public, breached, researched or supplied by someone else. Use a pre-agreed safe word or a separate trusted contact route for urgent requests.
Is all AI-generated content deceptive?
No. Synthetic content has legitimate uses. The risk comes from undisclosed or misleading use inside a situation designed to influence belief or action.
If your identity or media is being impersonated
Save the original evidence, report the account or content to the platform, warn affected contacts independently and contact the relevant bank, employer or authority if money, access or confidential information is at risk.
Scamology Insights
Separate what AI changes from what the scam still requires
These apps, tools, books and courses can help you look beyond a convincing voice, image or video. They show where a check made through contact details you found yourself can interrupt the scam. Links are intentionally withheld in this production version.
Continue understanding
Explore related guidance
Evidence and currency