Press ESC to close

Deepfake scams: How to verify a suspicious voice, video or photo

A familiar voice asks for emergency money. A senior executive appears on a video call approving a payment. A shocking clip seems to show a public figure saying something extraordinary. All three could be convincing, manipulated or simply presented out of context.

Treat suspicious media as an unverified claim, not as proof. Pause any payment or disclosure, contact the supposed sender through a channel you already trust, and investigate where the recording originated. Visual glitches can raise suspicion, but provenance and independent confirmation are stronger evidence. AI detectors can help with screening, but their results should never decide the matter alone.

The safest response is to verify the story, not diagnose the technology

You rarely need to establish whether a call or clip was technically produced by artificial intelligence. Your immediate task is to determine whether the person, event and request are genuine.

If a message demands money, credentials or confidential information:

  1. Stop the interaction. Do not let the caller keep you under pressure while you make a decision.
  2. Use a separate channel. Call the person on a number already saved in your contacts, message them through a previously established account or contact someone close to them.
  3. Find official contact details independently. If the caller claims to represent a bank, employer, school or government agency, do not use a number or link supplied in the suspicious message.
  4. Verify the request itself. Ask whether the emergency, payment instruction or account change is real.
  5. Refuse unusual payment methods. Requests for gift-card codes, cryptocurrency or an immediate transfer should receive extra scrutiny.

The US Federal Trade Commission advises people receiving apparent family-emergency calls not to trust the voice alone and to call the relative using a number they know. The FBI similarly recommends independently researching and contacting the person or organisation being impersonated. ([consumer.ftc.gov](https://consumer.ftc.gov/consumer-alerts/2023/03/scammers-use-ai-enhance-their-family-emergency-schemes?utm_source=openai))

How to handle a suspected voice-cloning call

A cloned voice may reproduce recognisable pitch and speech patterns. Trying to decide whether it “sounds artificial” is therefore less reliable than checking the caller’s identity and claims.

Warning signs often come from the behaviour surrounding the call:

  • an extreme emergency that supposedly requires immediate action;
  • instructions not to contact anyone else;
  • a demand for money, passwords, verification codes or financial details;
  • a new phone number or messaging account;
  • a supposed lawyer, police officer or doctor taking over the conversation;
  • resistance when you suggest ending the call and calling back independently.

Hang up and initiate your own call. If the person cannot be reached, contact another relative, colleague or relevant institution. Do not call a replacement number dictated by the suspicious caller.

Families can also agree on a private verification phrase. Choose something that is not published online or easily inferred from birthdays, pets, schools and other social-media information. A safe phrase is an additional check, not a substitute for calling back through a trusted channel.

For workplaces, payment changes and requests for sensitive data should follow a documented approval process. A convincing voice note or video meeting should not override callback procedures, two-person approval or established supplier details. The FBI recommends independently verifying communications involving financial transactions and using multi-factor authentication to reduce account risk. ([fbi.gov](https://www.fbi.gov/contact-us/field-offices/sanfrancisco/news/fbi-warns-of-increasing-threat-of-cyber-criminals-utilizing-artificial-intelligence?utm_source=openai))

How to investigate a suspicious video

Find the earliest reliable source

A reposted clip may have lost its date, caption and surrounding footage. Look for:

  • the original publication or the earliest version you can find;
  • a complete interview, speech, livestream or broadcast;
  • publication on the verified website or established account of the organisation involved;
  • a clear date, location and description of the event;
  • independent reporting that confirms the same event or statement.

Do not assume that a genuine recording is being presented honestly. A real clip can be cropped, slowed down, paired with different audio or attached to a false claim. CISA distinguishes deepfakes from simpler manipulations such as authentic footage presented out of context and recommends checking multiple sources. ([cisa.gov](https://www.cisa.gov/sites/default/files/publications/mdm-inauthentic-content-product-english_508.pdf?utm_source=openai))

Search individual frames

Pause the video on a clear, distinctive frame and use reverse-image or visual search tools. An older match may reveal that the footage came from another event, country or year. Repeat the process with several frames because added captions, cropping and compression can affect the results.

Treat visual defects as clues, not verdicts

Lip-sync problems, inconsistent lighting, distorted accessories, unnatural movement and irregular background details may justify closer examination. However, weak connections, video compression, filters and editing can also create strange artefacts in authentic footage.

Conversely, the absence of obvious defects does not establish authenticity. When the evidence remains inconclusive, say that the video could not be verified rather than declaring it fake or genuine.

How to check a suspicious photograph

Start with the claim attached to the picture: who supposedly took it, when and where was it captured, and where was it first published?

Then:

  • run a reverse-image search using the full image and significant crops;
  • look for an original version from a photographer, news organisation or institution;
  • compare signs, buildings, weather, clothing and shadows with the claimed place and time;
  • check whether reliable sources document the event;
  • inspect abrupt edges, repeated textures and inconsistent reflections, while remembering that these are only warning signs.

Metadata can sometimes provide useful context, but it is not decisive. Social platforms may remove it, editing software may change it and metadata does not prove that the scene in front of the camera was genuine. Provenance systems and embedded credentials can strengthen an authenticity assessment when present and valid, but missing provenance does not automatically mean that an image is fake. NIST describes provenance, watermarking and detection as evolving technical approaches rather than universal solutions. ([nist.gov](https://www.nist.gov/publications/reducing-risks-posed-synthetic-content-overview-technical-approaches-digital-content?utm_source=openai))

Why an AI detector should not be the final judge

A detector normally returns a score or classification based on patterns learned from test material. Its performance can change with the type of media, the generation method, compression, cropping and later editing.

That creates two important risks:

  • False positives: authentic material may be classified as synthetic.
  • False negatives: manipulated material may pass as authentic.

Even professional identity-verification guidance requires detection systems to be tested for both types of error and recommends supplementing automated analysis with manual review. NIST also notes that synthetic-content transparency and detection methods remain relatively new and require continued evaluation. ([pages.nist.gov](https://pages.nist.gov/800-63-4/sp800-63a.html?utm_source=openai))

Use a detector as one supporting signal. For a consequential decision, combine its result with the original source, contextual evidence, independent reporting and direct confirmation from the person or organisation concerned.

A label or watermark should be treated in the same way. Its presence may disclose that content was generated or edited, but it does not automatically make the content fraudulent. Its absence is not proof that a recording is authentic.

What to do if you already sent money or disclosed information

Act quickly rather than waiting to prove that a deepfake was involved.

If you sent money

  • Contact your bank, card issuer, payment service or cryptocurrency platform through its official channel.
  • Explain that the transaction resulted from impersonation or fraud and ask whether it can be stopped, recalled or flagged.
  • Do not pay a supposed recovery service that guarantees it can retrieve the money.
  • Report the incident to the appropriate fraud-reporting service or police authority in your country.

If you exposed an account

  • Change the password from a trusted device.
  • End unfamiliar sessions and remove unknown recovery details or connected applications.
  • Enable multi-factor authentication where available.
  • If the password was reused, change it on every other affected account.
  • Contact the organisation immediately if you disclosed a one-time code, banking credential or workplace secret.

Preserve useful evidence

Save messages, usernames, phone numbers, email headers, transaction references, links and screenshots. Record the time of each contact and what the scammer requested. Preserve the material without reposting it publicly, especially if it is defamatory, intimate or designed to spread panic.

A quick verification checklist

Before believing or acting on suspicious media, ask:

  • Who first published it?
  • Can I find the complete original rather than a reposted extract?
  • Is the claimed person or organisation confirming it through an established channel?
  • Do independent and credible sources support the claim?
  • Is someone using urgency, fear or secrecy to prevent verification?
  • Am I being asked for money, credentials or an exception to normal procedure?
  • Have I contacted the supposed sender using details I already trusted?

The most useful rule is simple: the more emotionally urgent a message feels, the more important independent verification becomes. A voice, face or photograph can no longer serve as sufficient proof of identity on its own.

Sources

  1. Senior U.S. Officials Continue To Be Impersonated in Malicious Messaging Campaign — FBI — https://www.fbi.gov/investigate/cyber/alerts/2025/senior-us-officials-continue-to-be-impersonated-in-malicious-messaging-campaign
  2. https://www.fbi.gov/investigate/counterintelligence/emerging-and-advanced-technology/super-intelligence
  3. Tools of Disinformation: Inauthentic Content — CISA — https://www.cisa.gov/sites/default/files/publications/mdm-inauthentic-content-product-english_508.pdf
  4. NIST Special Publication 800-63A — https://pages.nist.gov/800-63-4/sp800-63a.html
  5. Scammers use AI to enhance their family emergency schemes — Federal Trade Commission — https://consumer.ftc.gov/consumer-alerts/2023/03/scammers-use-ai-enhance-their-family-emergency-schemes
  6. https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/04/approaches-address-ai-enabled-voice-cloning?page=0
  7. Reducing Risks Posed by Synthetic Content — NIST — https://www.nist.gov/publications/reducing-risks-posed-synthetic-content-overview-technical-approaches-digital-content
  8. https://consumer.ftc.gov/consumer-alerts/2024/04/fighting-back-against-harmful-voice-cloning
  9. FBI Warns of Increasing Threat of Cyber Criminals Utilizing Artificial Intelligence — https://www.fbi.gov/contact-us/field-offices/sanfrancisco/news/fbi-warns-of-increasing-threat-of-cyber-criminals-utilizing-artificial-intelligence
  10. https://mfc.nist.gov/
  11. https://consumer.ftc.gov/articles/scammers-use-fake-emergencies-steal-your-money
  12. https://www.nist.gov/system/files/documents/2024/02/15/ID016%20-%20Computing%20Research%20Association%20Computing%20Community%20Consortium.PDF
  13. https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=959123
  14. https://www.nist.gov/system/files/documents/2025/09/29/FINAL_PRESO_DIG_V4_For%20DISTRO.pdf
  15. https://www.nist.gov/programs-projects/guardians-forensic-evidence
  16. https://www.nist.gov/system/files/documents/2025/05/08/IHS_PosterorBrochure_attachment.pdf
  17. https://www.nist.gov/news-events/news/2025/08/nist-guidelines-can-help-organizations-detect-face-photo-morphs-deter
  18. https://www.ftc.gov/system/files/ftc_gov/pdf/DeFake-Abstract.pdf
  19. Scammers use AI to enhance their family emergency schemes | Consumer Advice — https://consumer.ftc.gov/consumer-alerts/2023/03/scammers-use-ai-enhance-their-family-emergency-schemes?utm_source=openai
  20. FBI Warns of Increasing Threat of Cyber Criminals Utilizing Artificial Intelligence — FBI — https://www.fbi.gov/contact-us/field-offices/sanfrancisco/news/fbi-warns-of-increasing-threat-of-cyber-criminals-utilizing-artificial-intelligence?utm_source=openai

Gabriel Čatár

I'm interested in everyday things we encounter but don't always understand. From technology and personal finance to practical guides and fascinating facts about the world. At Offpitch, I focus on topics worth exploring in greater depth, checking the facts and explaining them in a clear, accessible way. My goal is to make sure readers take away something useful from every article.