A deepfake is synthetic media — video, audio or images — created or altered with AI to show something that never happened, usually involving real people. The term emerged in 2017; since then, tools have become cheap, fast and convincing enough that anyone with a laptop can produce realistic fakes.

How deepfakes are made

  • Face swapping: replacing one person’s face with another’s in video.
  • Lip-sync and reenactment: making someone appear to say words they never said.
  • Voice cloning: generating speech in a person’s voice from short samples.
  • Fully synthetic images and video: generated from text prompts by image and video models.

Modern generative models need far less source material than early techniques; a few seconds of audio can be enough to clone a voice convincingly.

How they are misused

  • Fraud and scams: cloned voices of relatives asking for urgent money, as the US Federal Trade Commission has warned; fake video calls impersonating executives to authorize transfers.
  • Non-consensual intimate imagery: the most widespread harmful use, overwhelmingly targeting women, now addressed by laws in many countries.
  • Political disinformation: fake statements or robocalls attributed to candidates and officials.
  • Harassment and reputational attacks.
  • The “liar’s dividend”: real evidence dismissed as fake because deepfakes exist.

How to spot them — and the limits

Visual clues can include unnatural blinking, mismatched lighting, blurry edges around the face, odd hands or teeth, inconsistent reflections, or audio slightly out of sync. But generation quality improves constantly, and these clues are becoming unreliable. Automated detectors exist but can be fooled and produce errors.

More reliable strategies focus on context and verification:

  1. Check the source: who published it first? Is it from an official account or reputable outlet?
  2. Look for corroboration: would a statement this significant be reported elsewhere?
  3. Reverse-search images and key frames.
  4. Be suspicious of urgency and emotion — scams rely on pressure.
  5. Check provenance information such as Content Credentials, when available — see our explainer on content provenance.

Protecting yourself and your organization

  • Agree on a family code word for emergencies to defeat voice-clone scams.
  • Verify unusual requests — especially for money or credentials — through a separate, known channel.
  • Require multiple approvals for large payments in companies.
  • Limit public voice and video samples where practical, and use privacy settings.
  • Report non-consensual deepfakes to platforms and, where applicable, to authorities.

Laws increasingly target harmful deepfakes: criminalizing non-consensual intimate deepfakes, regulating deceptive election content and requiring disclosure. The EU AI Act requires deployers to disclose deepfakes, with exceptions for evidently artistic or satirical works.

If you find a deepfake of yourself

  1. Save evidence: screenshots, links, dates and accounts that shared it.
  2. Report it to the platform using its reporting tools for impersonation or non-consensual intimate imagery; many platforms prioritize such reports.
  3. Use removal services: some organizations help victims of intimate image abuse get content removed across platforms.
  4. Contact the police if it involves threats, extortion or intimate images — creating or sharing such content is a crime in many jurisdictions.
  5. Seek support from trusted people and specialized helplines.

Related: content provenance and C2PA and AI scam calls and texts.

Bottom line

Verify the source, not just the pixels.

What to watch next

  • Adoption of content provenance standards by cameras, platforms and AI generators.
  • New laws on non-consensual deepfakes and election deepfakes.
  • Advances and limits of detection technology.
  • Platform policies for labelling synthetic media.

Key takeaways

  • Deepfakes are AI-generated or altered media depicting things that never happened.
  • Voice cloning now needs only short audio samples, fueling scams.
  • The most common harmful use is non-consensual intimate imagery.
  • Visual clues and detectors are unreliable; source verification and provenance matter more.
  • Family code words and out-of-band verification defeat many deepfake scams.

Frequently asked questions

Are deepfakes illegal?

Creating synthetic media is not illegal in itself, but many uses are: fraud, defamation, harassment, non-consensual intimate imagery and, in some places, deceptive election content.

Can AI detectors reliably identify deepfakes?

Not reliably on their own. They can help, but they miss new techniques and produce false positives.

What should I do if someone makes a deepfake of me?

Document it, report it to the platform, seek support from relevant helplines and, depending on your country and the content, report it to the police.