Deepfakes are synthetic media, video, audio or images, created or altered with AI to show someone saying or doing something they did not. A few years ago they were often easy to spot. Today, the best ones can fool most people, and the tools to make them are widely available.
Updated September 2026: we added the details and enforcement date of the US TAKE IT DOWN Act, when the EU AI Act’s deepfake disclosure rules apply, and official warnings about voice clone scams.
The main types
- Face swaps and synthetic video that put one person’s face on another’s body or generate a scene from scratch
- Voice clones that mimic a real person’s voice from a short sample
- Generated images of events that never happened
- Edited real media, where a genuine clip is altered to change what someone said
Why the old tells are fading
Guides used to advise looking for odd blinking, mismatched earrings, warped backgrounds or extra fingers. These glitches still appear sometimes, but newer models make fewer of them. Relying on visual tells alone now gives false confidence, both ways: you may miss a good fake and dismiss a real video that happens to look odd.
Habits that work better than squinting
- Check the source. Who posted it first? Is it from a verified, accountable outlet or an anonymous account?
- Look for corroboration. Real news events are usually covered by multiple reliable organizations. If something shocking appears in only one place, wait.
- Search for the original. A reverse image search or searching key phrases can reveal older or unaltered versions.
- Notice emotional pressure. Content designed to make you instantly angry or afraid deserves extra suspicion.
- Check for content credentials. Some cameras, editing tools and AI generators attach provenance information, based on the C2PA standard and known as Content Credentials, showing how an image was made. Its absence proves nothing, but its presence can help.
Voice clone scams
One of the most damaging uses is fraud. Criminals clone a voice from social media clips and call a relative or colleague claiming an emergency. Police and consumer protection agencies in several countries have warned about these scams. In the US, the Federal Trade Commission warned in 2023 that a scammer needs only a short audio clip to clone a family member’s voice, and the FBI has advised families to agree on a secret word or phrase.
Protect your family Agree on a private code word with close family members to use in any urgent request for money. If you get a panicked call, hang up and call the person back on a number you already know.
What platforms and lawmakers are doing
- Labeling. Major platforms have introduced labels for AI-generated content and require creators to disclose realistic synthetic media in some cases. YouTube, for example, has required that disclosure since March 2024.
- Watermarking and provenance. AI companies increasingly embed invisible watermarks, such as Google DeepMind’s SynthID, or C2PA metadata in generated content, though these can sometimes be stripped.
- Laws. The EU AI Act requires disclosure of deepfakes, a transparency duty that applies from August 2026, as covered in our AI Act explainer. In the United States, the TAKE IT DOWN Act, signed in May 2025, targets non-consensual intimate imagery, including AI-generated "digital forgeries." Since May 2026, the FTC has enforced its requirement that platforms remove a reported image, and known copies, within 48 hours, and many states have passed laws on election deepfakes and intimate imagery.
If you are targeted
If a deepfake of you circulates, document it with screenshots and links, report it to the platform using its impersonation or non-consensual imagery process, and seek legal advice where appropriate. Organizations that support victims of image-based abuse can help with takedowns.
Process beats perception
You cannot out-squint a good deepfake. What protects you is process: check the source, look for corroboration, slow down when content is designed to provoke, and verify urgent requests through a channel you trust.
Sources
- Take It Down Act enforcement starts now, Federal Trade Commission, May 2026
- Scammers use AI to enhance their family emergency schemes, Federal Trade Commission, March 2023
- Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud, FBI Internet Crime Complaint Center, December 2024
- Content Credentials, C2PA
- SynthID, Google DeepMind
- How we’re helping creators disclose altered or synthetic content, YouTube, March 2024
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), EUR-Lex
