The generative AI boom has opened Pandora’s box, and deepfakes are just one of its many consequences. You no longer need a movie studio or research lab to create convincing fake videos, images, and voice clips.

So, what are deepfakes, and are they dangerous? They certainly can be. While some are harmless—like clips of US presidents playing Minecraft together—others are used to spread misinformation, commit fraud, target people with harassment, or even create explicit scenes without the consent of the people in them.

What's in this article?

Worse yet, it’s getting harder to distinguish fake content as the technology evolves. Realistically, you can’t fact-check every video, image, or voice clip you come across, which makes deepfakes much easier to spread.

Below, we’ll cover how deepfakes are created, the most common uses and attacks involving them, and practical tips to tell them apart from real content.

What are deepfakes and how are they created?

Deepfakes are videos, images, or audio edited or generated with AI to make someone appear to say or do something they never actually did. The term combines “deep learning,” a machine learning technique that finds patterns in large amounts of data, with the word “fake.”

Before it can generate convincing content, the AI is trained on large collections of photos, videos, or voice recordings. By analyzing enough examples, it learns how a person’s face moves, how their voice sounds, and how they speak or react in different situations.

Once the training is complete, the AI model can copy a person’s face, voice, or expressions with surprising accuracy. It can then generate new content or replace someone’s likeness in existing media, often making the result difficult to spot at first glance.

What are deepfakes used for?

Deepfakes are used for everything from internet jokes and movie effects to AI-driven identity theft scams, political misinformation, and non-consensual intimate images (NCII). Their purpose depends entirely on the person creating them and how they choose to share them.

While a few researchers are exploring limited therapeutic uses, most people encounter deepfakes through social media, messaging apps, or online forums. As the technology improves, creating convincing fake content also becomes easier and more accessible.

Why are deepfakes dangerous?

Deepfakes are dangerous because they can spread false information much faster than it can be checked or corrected. A convincing fake video or voice recording can damage reputations, influence public opinion, or trick people into making costly mistakes.

They also make AI scams more believable by cloning someone’s voice or face to impersonate them. At the same time, victims often have little control over where deepfakes are posted or how quickly they spread across the internet.

A recent example involved Grok’s image generation tool on X/Twitter. Users abused the feature to create non-consensual sexual deepfakes, including some involving minors. Rather than addressing the underlying safety problem, X initially responded by restricting image generation to paying subscribers.

The backlash was serious enough for Ofcom to investigate X’s compliance with the UK Online Safety Act. In extreme cases, the law allows the regulator to seek court orders that block platforms in the UK if they repeatedly fail to deal with illegal content.

French prosecutors have also launched an investigation into X, and the company is facing multiple lawsuits in the US (see, for example, this case and this one) alleging Grok was used to create child sexual abuse material (CSAM).

Common deepfake attacks

Most deepfake attacks rely on the same idea: convincing you that fake content is real. Whether the goal is fraud, misinformation, or harassment, here are some of the most common ways attackers use the technology:

  • Voice cloning attacks: Someone may clone a family member’s, friend’s, or CEO’s voice to ask for money or sensitive information. Others use voice clones to impersonate classmates or coworkers as a form of harassment.
  • Business email compromise (BEC): Attackers combine phishing with AI voice cloning to convince employees to transfer funds or reveal company data.
  • Bypassing identity checks: Criminals can use face swaps and voice cloning to fool selfie verification, identity checks, or call center authentication.
  • Fake endorsements: Celebrities and public figures may appear to promote investment and cryptocurrency scams or miracle products they never supported.
  • Brand impersonation: Attackers posing as customer support agents or company representatives can trick people into revealing credentials or sending payments.

How to detect deepfakes

While it’s becoming harder to tell what’s real and what’s not, a few clues can still help you identify deepfakes:

  • Watch for unnatural movements: Look for odd facial expressions, stiff head movements, or lip movements that don’t quite match the audio. People in deepfake videos tend to face the camera and don’t move much.
  • Check the lighting: Shadows, reflections, and skin tones may change unexpectedly or look inconsistent.
  • Listen for voice issues: AI-cloned voices can sound flat, “metallic” or compressed (like if you exported an mp3 file at 96 kbps), rushed, or have strange pauses and pronunciation.
  • Look at the background: Objects, text, or people in the background may warp, flicker, or change between frames. Text, in particular, tends to look like gibberish.
  • Verify the source: Check whether trusted news outlets or the original creator have shared the same content.
  • Reverse-search key frames: A reverse image search can sometimes reveal the original image or video used to create the deepfake.
  • Use AI detection tools with caution: Some services can estimate whether a video was manipulated, but none are completely reliable. Treat their results as one clue rather than proof.

How to stay safe against deepfakes

No method offers complete protection, but a few simple habits can reduce the chances of becoming a victim or falling for a convincing fake:

  • Verify before sharing: Before reposting a viral video of a politician making outrageous claims, a celebrity caught in a scandal, or a CEO announcing unexpected news, check whether reputable sources have confirmed it’s genuine.
  • Limit what you post publicly: Fewer public photos and videos give bad actors less material to work with. Set your social media accounts to private and trim your friend list to people you know and trust.
  • Use strong account security: Choose a passphrase instead of a password and enable two-factor authentication to make it harder for someone to hijack your accounts.
  • Double-check unusual requests: If someone calls or sends a voice message asking for money or sensitive information, contact them through another method first. This can help you avoid scams sent from a compromised account or a convincing voice clone.
  • Agree on a verification phrase: Choose a private word or phrase that only trusted people know. If someone makes an urgent request over the phone or in a voice message, ask them for it before taking action. Keep it private and change it if you think it has been exposed or shared through text or social media.
  • Report harmful deepfakes: Report fake or non-consensual content to the platform hosting it and, where appropriate, to the relevant authorities.
  • Stay informed: Deepfake technology changes quickly, so keeping up with new scams and common warning signs makes them easier to spot.

What are deepfakes? FAQs

Are deepfakes illegal?

Deepfakes are typically only illegal when used in harmful or unlawful ways. In the US, laws such as the Take It Down Act target NCII, including AI-generated ones, while section 138 of the Data (Use and Access) Act 2025 introduced similar protections in the UK.

Of course, none of this is legal advice. Consult your local legislation if you’re unsure about the legality of deepfakes.

Why are deepfakes unethical?

Deepfakes are considered unethical because they can make people appear to say or do things that they never said or did. They can damage someone’s reputation, spread false information, be used in scams, and violate peoples’ privacy without their knowledge or consent.

How can you find deepfakes of yourself?

Finding deepfakes of yourself is difficult because there isn’t a reliable scanner for them. Google Images also limits reverse image results for real people to protect privacy. Bing and Yandex are pretty hit-or-miss, and paywalled tools like PimEye aren’t much better.

Moreover, people generating non-consensual deepfakes don’t always post them publicly, so it would be impossible to find out if someone is using your photos. The safest approach is to set your socials to private and limit what you post online so you don’t give bad actors more ammo.

What are the benefits of deepfakes?

The potential benefits of deepfakes come with severe caveats. Researchers are exploring their use in therapy, where consensually obtained deepfakes could help patients process trauma or grief under professional supervision. Deepfake therapy is still experimental, and its ethical and legal concerns need careful study.

What are some examples of deepfakes in media?

Deepfakes in media are commonly used in movies and TV shows, often for de-aging actors or digitally recreating characters. One example is the appearance of a certain spoiler-y character in the season 2 finale of The Mandalorian.

They also appear in music videos, such as Kendrick Lamar’s The Heart Part 5, where his face transforms into several public figures like Will Smith and Kobe Bryant.

Read more: