What a deepfake is
A deepfake is audio, video or an image that has been generated or altered by software so that it appears to show a real person saying or doing something they did not. The word combines “deep learning,” a family of techniques within machine learning, with “fake.”
At a general level, these systems are trained on large collections of images, video or recordings, and learn the patterns of a face, a voice or a style of speech. Once trained, they can produce new material that resembles what they learned. This is a specific use of generative AI, and the quality of the results depends heavily on the training data and the tools involved. This article does not explain how to make them. It focuses on why they matter and how to respond.
Not all synthetic media is harmful. It is used in film production, accessibility tools, language dubbing and satire that is clearly labeled. The concern is deception, or use without a person’s consent.
Main kinds of harm
Non-consensual intimate imagery. One of the most serious and well-documented harms is the creation of sexual images or videos of real people without their agreement. Victims can experience distress, damage to reputation and harassment. Many places have introduced or are considering laws on this, and rules vary widely.
Fraud and impersonation. A cloned voice or a manipulated video can be used to impersonate a relative, an executive or an official in order to request money or sensitive information. This is a modern variant of social engineering, and it targets trust rather than technical weaknesses. Older relatives can be especially exposed, as described in the discussion of elder scams.
Disinformation. Fabricated clips of public figures can be used to mislead audiences, particularly when they spread quickly before anyone checks them. They can feed wider patterns of online misinformation. Researchers also point to a second effect: once fakes are widely known, real recordings can be dismissed as fake, which erodes confidence in genuine evidence.
Harassment and reputational damage. Fabricated material can be used to embarrass, intimidate or discredit private individuals, including students and coworkers, as well as public figures.
How common each of these harms is, and how effective the fakes are, is hard to measure reliably, and claims about scale should be treated with caution.
Why detection is difficult
Quality has improved, and there is an ongoing contest between tools that generate media and tools that try to detect it. Automated detectors can be wrong in both directions, and results can change as generation methods change. No single check is dependable on its own. That is why most advice stresses habits of verification over trust in a particular tool.
Signs that something may be synthetic
These cues sometimes appear, but their absence does not prove authenticity:
- Odd details in hands, teeth, jewelry, text, reflections or backgrounds.
- Lighting or shadows that do not match across the scene.
- Lip movements slightly out of step with the audio, or an unusually flat voice rhythm.
- Blurring or flickering around the edges of a face or hairline.
- A claim that seems designed to provoke an immediate reaction.
Context is often more useful than visual detail. Ask who first published it, whether reliable outlets report the same thing, and whether the original source can be found. The broader topic of synthetic media awareness covers more general cues, and standard fact-checking methods such as reverse image searches and looking for the earliest version still apply.
What to do if you encounter one
- Pause before sharing. Sharing a doubtful clip, even to criticize it, can help it spread.
- Check through a second channel. If a call or message from a relative or manager asks for money or secrets, contact them using a number or method you already trust. Some families and teams agree on a private check phrase for this purpose.
- Save evidence, such as links and screenshots, before reporting.
- Report it to the platform. Most large services have reporting routes for impersonation and intimate imagery, though speed and outcomes vary. The way content moderation works helps explain why responses differ.
- Contact local authorities or legal advice services where fraud, threats or intimate imagery are involved. Laws and available remedies differ by place, and this article is general information, not legal advice.
If you or someone you know has been targeted and is distressed, consider reaching out to a trusted person or a qualified professional, and contact local emergency or crisis services if there is any immediate danger.
What organizations and policymakers are weighing
Several approaches are discussed, each with trade-offs: labeling or watermarking synthetic content, rules requiring disclosure, platform policies on manipulated media, criminal and civil laws for specific harms, and training staff to verify unusual requests. Questions about who bears responsibility are part of the larger debate over platform liability. Labeling can be bypassed, and broad bans risk catching legitimate uses such as satire, so there is no agreed single answer.
A realistic stance
The sensible response is neither panic nor dismissal. Deepfakes are one tool among many for deception, and the habits that protect against older scams still help: slow down, verify through independent channels, and be wary of content that asks for urgent action or strong emotion.