Deepfake: What It Is, How to Spot It, and Legal Consequences

👁 115 reads

A deepfake is a video, photo, or voice recording generated or altered by artificial intelligence to show a real person doing or saying something they never did. The word emerged in 2017 from combining “deep learning” and “fake,” when a Reddit user with that pseudonym began publishing face-swapped videos of actresses. Nine years later, you can create the same effect in a minute on your phone, and the victims are no longer just celebrities but ordinary people: someone receives a voice message “from their daughter” asking for money, or sees an ad where a famous doctor recommends a miracle cure. Below: how it works, how to spot a deepfake, legal consequences, and what to do if you’re targeted.

How Deepfakes Are Made and Why They’re So Easy

Until 2023, you needed hundreds of photos of the victim and an expensive graphics card. Today, one photo and a few seconds of voice are enough: image models transfer a face onto any body and background, voice models clone timbre from a short sample, and video generators animate a photo into a moving, speaking figure. The same tools we cover in guides about editing photos with AI and AI voice-over become deepfake factories in a scammer’s hands. The only difference is whether the person in the photo consented.

Three Types of Deepfakes You’ll Encounter

Investment scams with celebrities. Social media ads where a well-known presenter or politician promotes an “investment platform.” Financial regulators have warned about this scheme for years; money sent to these platforms never comes back.

Voice impersonation fraud targeting families. A phone call or voice message from a “child” or “grandchild” in trouble. Voice clones are made from TikTok recordings or Instagram stories. A family security password defeats this scheme instantly.

Intimate and political deepfakes. Face-swapped pornographic material, most often targeting women, and fabricated statements by politicians before elections. We covered the latter in our analysis of deepfakes in 2026 politics, and the market for trading faces in deepfake for sale.

How to Spot a Deepfake: Six Things We Check

First, the source: does the video come from the person’s official channel or newsroom, or from an account created a week ago. Second, mouth and teeth: during speech, synchronization can be a fraction of a second off, and teeth look like a uniform strip. Third, face boundaries: the hairline, ears, glasses, and neck are places where the swap “glues together.” Fourth, hands and jewelry, which change shape between frames. Fifth, voice: lack of breaths, even rhythm, strange accents in names. Sixth, context: if a famous person promises profit, medicine, or asks for a transfer, it’s a scam regardless of image quality. None of these features alone is sufficient, because the best 2026 models no longer have obvious errors; it’s the sum that counts.

Tools for Verification

Images and videos from Google tools have an invisible SynthID watermark that can be verified; C2PA metadata added by some cameras and software works similarly. However, this is only proof in one direction: the absence of a mark doesn’t mean the material is authentic, because it can be removed through a screenshot. Reverse image search in Google shows whether a photo circulated earlier in a different context. AI detectors work better on text than on video, as we explain in our article about AI detectors.

Legal Consequences of Deepfakes and What the AI Act Says

Laws in most jurisdictions don’t have a separate “deepfake” crime, but prosecute the consequences: fraud, defamation, image rights violations, threats, and for intimate materials without consent, laws about recording and distributing nude images. The EU AI Act, effective August 2, 2026, imposes an obligation to clearly label AI-generated content, including deepfakes, which we covered in our article about labeling synthetic media under the AI Act. This is commentary, not legal advice: for a specific case, consult a lawyer or law enforcement with preserved evidence.

What to Do If a Deepfake Targets You

Take screenshots with visible URLs and dates, save links and account identifiers. Report the material on the platform as impersonation (Meta, TikTok, YouTube, and X have separate forms for deepfakes). File a report with law enforcement, and for financial fraud, notify your bank and financial regulators. Inform your contacts and followers that the material is fake before someone else does it for you.

How to Report a Deepfake on Specific Platforms

Step-by-step instructions: Facebook and Instagram, TikTok, and YouTube.

Frequently Asked Questions

Is creating deepfakes legal?

Using the tool itself is legal; it becomes a crime when someone’s image or voice is used without consent in a way that deceives, defames, or violates privacy. Satire and parody with clear labeling are in a different situation than impersonation.

Can you recognize a deepfake with 100 percent certainty?

No. Detectors provide probability, and the best 2026 models regularly fool them. The most reliable method remains checking the source and contacting the person through a different channel.

How can you protect your family from voice scams?

Agree on a password that only family members know, and a rule that money requests are confirmed by calling back a known number. This works better than any detector.

Sources verified September 17, 2026: financial regulator communications about scams using public figures’ images, AI Act text (Regulation 2024/1689, Article 50 on transparency) and its implementation timeline, SynthID and C2PA documentation, our own tool tests described in linked articles.

// read moreDeepfakes: How to Defend Against AI-Generated Fake VideosMagazineHow to Report a Deepfake on Facebook and InstagramGuidesAI Act Deepfake Rules: Who Must Label AI Content From Aug 2026Guides

Kinetyra