What Are Deepfakes? Laws, Harms, and Spotting Tips

What are deepfakes? Deepfakes are AI-made images, videos, or audio that make it look like a real person did or said something they never did. In plain terms, the deepfake meaning is “synthetic media that can be hard to tell from reality,” which is why deepfake AI can be used for scams, harassment, and non-consensual sexual content.

• What are deepfakes? (Deepfakes definition)
• How does deepfake AI work?
• Why deepfake technology matters: harms and risks
• Real-world deepfake examples (including Taylor Swift)
• Deepfake laws: are deepfakes illegal?
• How to spot a deepfake (quick checklist)
• What to do if you’re a victim of a deepfake
• Which deepfake guide do you actually need?
• FAQ: deepfake meaning, legality, and spotting

What are deepfakes? (Deepfakes definition)

A deepfake is a synthetic image, video, or audio recording created by AI that depicts someone doing or saying something they never did. The key point in the deepfakes definition is that it uses machine-learning systems to imitate a real person’s face, voice, or movements so convincingly that the result can be difficult to distinguish from genuine footage.

People also search for this as deep fake (two words) or “what is a deepfake.” It all refers to the same idea: AI-generated media that can be used to mislead, embarrass, or harm someone—especially when it’s shared widely online.

Deepfake meaning in one sentence

Deepfake meaning: AI-made media that convincingly imitates a real person’s appearance or voice, often to trick viewers or target someone.

What deepfakes are (and aren’t)

• Deepfakes are: AI-generated or AI-edited audio/video/images that put realistic words/actions on a person who didn’t actually do them.
• Deepfakes aren’t automatically illegal or malicious: The same underlying methods can be used for satire, film effects, or accessibility tools. The harm usually depends on consent, context, and intent.

How does deepfake AI work?

At a high level, deepfake technology works by training AI systems on lots of examples of a person’s face or voice. The system learns patterns—like how a person smiles, turns their head, or pronounces certain sounds—then generates new media that imitates those patterns.

You don’t need to understand the math to grasp the risk: once the AI has enough examples, it can produce a convincing “performance” that never happened. And because the output can look like “real video,” deepfakes can hijack the credibility we normally give to recorded evidence.

Why deepfakes can look “real” so fast

• Easy access: Tools that generate images and video have become easier to use, lowering the barrier for harassment and fraud.
• Social sharing: Even a fake that’s only “pretty convincing” can go viral before anyone checks it.
• Compression and reposting: Re-uploads and low-quality copies can hide telltale artifacts.

Why deepfake technology matters: harms and risks

Deepfakes matter because they can cause real-world harm quickly—often before corrections catch up. They can ruin reputations, enable harassment, and make it harder for everyone to trust authentic evidence.

One of the clearest patterns in documented harms is that women and girls are the overwhelming majority of deepfake victims. Celebrities are disproportionately targeted, but ordinary people—including teenagers—are increasingly victimized.

Common harms linked to deepfakes

• Non-consensual sexual imagery (NCII): “Deepfake porn” that uses a real person’s face without consent.
• Scams and impersonation: Fake audio or video used to pressure someone into sending money or sensitive information.
• Blackmail and stalking: Synthetic media can be used to threaten exposure or humiliate someone.
• Trust damage: When deepfakes spread, people start doubting real footage too (“maybe everything is fake”).

Who is most at risk?

Based on the patterns described above, the highest-risk groups tend to be:

• Women and girls, especially targeted with sexual deepfakes.
• Public-facing people (teachers, nurses, local officials, creators) whose images are easy to collect online.
• Teens, who may be targeted by classmates and may not know where to report.

If you’re a parent or caregiver, you may also want Ban the Bots’ parent-focused resources at /parents/ for practical safety planning.

Real-world deepfake examples (including Taylor Swift)

Deepfakes aren’t a theoretical problem. A widely cited case involved Taylor Swift in January 2024, when AI-generated sexually explicit images spread across social media. In that incident, one post was viewed over 47 million times before removal—a clear example of how quickly deepfake content can travel and how long it can remain “up” before enforcement catches it.

Deepfake scams have also focused on celebrities: in November 2025, McAfee ranked Taylor Swift as the #1 celebrity most targeted by deepfake scammers. And in August 2025, Elon Musk’s Grok AI image tool reportedly generated sexually explicit deepfakes of Swift from innocuous prompts, reigniting debate over tool design and platform responsibility.

Why celebrity deepfakes matter to everyone

It’s tempting to think, “That’s a celebrity problem.” But celebrity cases show the underlying reality: if platforms can’t reliably stop a viral deepfake of one of the most recognizable people on Earth, it’s even harder for ordinary people—especially teens—who don’t have publicists, lawyers, or press attention.

For more incidents and context, see /ai-incidents/.

Deepfake laws: are deepfakes illegal?

Are deepfakes illegal? Sometimes—depending on what the deepfake depicts, whether the person consented, whether it’s intimate imagery, and which jurisdiction you’re in. The legal landscape is moving fast: as of spring 2026, 46 U.S. states have enacted laws targeting AI-generated synthetic media, often focusing on political ads, sexual content, or both.

Is deepfake porn illegal? (NCII and the TAKE IT DOWN Act)

A major U.S. federal change is the TAKE IT DOWN Act, which was signed into law by President Trump on May 19, 2025. It criminalizes the non-consensual publication of intimate images—whether the images are real or AI-generated—with penalties of up to 3 years in prison.

It also includes a platform obligation: covered platforms must remove flagged content within 48 hours of notice. Importantly, the law isn’t just symbolic—the first conviction under this law was issued in April 2026.

Can victims sue? The DEFIANCE Act (civil lawsuits)

Criminal law isn’t the only route. The DEFIANCE Act (Disrupt Explicit Forged Images and Non-Consensual Edits Act) passed the U.S. Senate unanimously in January 2026. It creates a federal right of action—meaning victims can sue creators and distributors—allowing statutory damages up to $150,000, or $250,000 if linked to sexual assault or stalking.

If you’re trying to understand how these issues are playing out in court, Ban the Bots tracks relevant cases at /ai-lawsuits/.

State deepfake laws and a Section 230 reality check

In addition to federal efforts, states have enacted their own deepfake laws. Many focus on election-related synthetic media, non-consensual intimate imagery, or both. Because state laws vary, what’s illegal in one state might be treated differently in another.

There’s also a real legal constraint in the U.S.: California AB 2655 (2024) was partially struck down by a federal judge citing Section 230, highlighting the ongoing tension between regulating online platforms and long-standing U.S. internet liability rules.

EU rules: labeling and takedowns

In the European Union, the approach includes transparency and platform accountability. The EU AI Act requires deepfakes to be labeled, and the Digital Services Act includes platform responsibilities that can require takedowns. If you want a broader primer on the EU framework, see /explainers/eu-ai-act.

Comparison: key deepfake-related rules at a glance

• U.S. federal (TAKE IT DOWN Act, 2025): Criminal penalties (up to 3 years) for publishing non-consensual intimate images (real or AI); platform takedown requirement within 48 hours after notice; first conviction April 2026.
• U.S. federal (DEFIANCE Act, Senate passed Jan 2026): Civil right to sue; statutory damages up to $150,000 (or $250,000 when linked to sexual assault or stalking).
• U.S. states (46 as of spring 2026): Mix of laws targeting synthetic media in political ads, sexual content, or both; details vary widely by state.
• EU (AI Act + Digital Services Act): Labeling requirement for deepfakes; platform obligations that support takedown and accountability processes.

How to spot a deepfake (quick checklist)

If you’re Googling how to spot deepfake content, here’s the practical truth: no single “tell” works every time. But many deepfakes still show small inconsistencies—especially when you slow down playback, zoom in, or compare with known-real clips.

How to spot a deepfake: 10 signals to check

1. Unnatural blinking (too little, too much, or oddly timed).
2. Lighting that doesn’t match across the face and background.
3. Blurry edges around hair or where the face meets the neck.
4. Weird teeth or tongue detail during speech.
5. Mouth movements that don’t sync with the audio.
6. Audio that sounds “flat” or lacks natural breathing and room tone.
7. Jittery frames during head turns or hand-to-face motion.
8. Inconsistent earrings/glasses that flicker or warp.
9. Too-perfect skin with smeared texture (like aggressive smoothing).
10. Context red flags: the clip appears with no original source, no full interview, or no reputable outlet confirming it.

Tools that can help detect deepfakes

When you need something beyond eyeballing, you can also use detection tools that are commonly referenced in deepfake discussions, such as Microsoft’s Video Authenticator or Sensity AI. Think of these like “second opinions”—helpful, but not perfect.

For a deeper guide on synthetic media tricks (and verification habits that actually work), Ban the Bots also has a broader explainer at /explainers/deepfakes.

What to do if you’re a victim of a deepfake

If you’re dealing with a deepfake of yourself or your child, focus on two goals: (1) preserve evidence, and (2) trigger takedown/reporting routes that have real leverage.

Step-by-step: what to do today

1. Document everything before it disappears: screenshots, URLs, usernames, timestamps, and where it was posted or shared.
2. Report on-platform using the service’s reporting tools (harassment, impersonation, NCII).
3. Use the federal takedown path when applicable: under the TAKE IT DOWN Act, platforms are required to remove flagged non-consensual intimate images (including AI-generated) within 48 hours of notice.
4. If the victim is a minor, report to NCMEC (National Center for Missing & Exploited Children). This is a key channel for child sexual exploitation material.
5. If the victim is an adult, consider contacting the Cyber Civil Rights Initiative, which is widely referenced for support around non-consensual intimate imagery.
6. Consider legal options: depending on the facts, the DEFIANCE Act framework (federal civil right of action, statutory damages) may matter, and state laws may also apply.

What not to do (common mistakes)

• Don’t negotiate with blackmailers or send more images “to prove it’s you.”
• Don’t assume one report is enough: save evidence and follow up, especially if the content is reposted.
• Don’t share the clip to “warn people” without thinking—reposting can amplify the harm.

If you’re trying to protect your workplace, school, or community

Deepfakes are now a routine part of harassment, fraud, and misinformation. Schools and employers can reduce harm by setting clear rules about AI-generated media and impersonation. If you’re drafting policy language, you can start from Ban the Bots templates like /no-ai-policy-template/ or /human-made-policy-template/.

Are deepfakes ever used for good? Legitimate uses

Not every use of deepfake technology is an attack. The same face- and voice-synthesis methods that power scams and non-consensual imagery are also used, with consent, for dubbing, accessibility, and film production — and a few real, well-documented cases show what "good" deepfake use actually looks like in practice.

Multilingual public-health and charity campaigns

In April 2019, the charity Malaria No More UK worked with AI video company Synthesia to produce a public-service video in which David Beckham appears to speak nine languages, each voiced by a real malaria survivor or doctor from that region. The 55-second spot was fully consensual — Beckham and Malaria No More commissioned it — and it reached over 800 million impressions and won a 2019 CogX award for "Outstanding Achievement in Social Good Use of AI." It's frequently cited as one of the first mainstream examples of synthetic media used to widen a message's reach rather than to deceive.

Restoring a voice lost to illness

After a 2014 tracheotomy tied to throat cancer left actor Val Kilmer largely unable to speak, AI voice company Sonantic built a synthetic version of his voice from archival recordings he supplied. That AI-generated voice was used, with Kilmer's participation, for his character's dialogue in Top Gun: Maverick (2022) — reported at the time by Fortune as voice-cloning technology restoring an ability the actor had lost, rather than impersonating him without consent (Paramount has said the released film's audio is Kilmer's own voice, so accounts of exactly how much of the AI model made the final cut vary). Sonantic was acquired by Spotify in 2022.

Film localization and dubbing

Traditional movie dubbing replaces the audio track but leaves the actor's original lip movements out of sync with the new language. Visual-dubbing companies like Flawless AI use deepfake-adjacent technology to adjust an actor's on-screen facial performance to match dubbed dialogue frame-by-frame, rather than generating a new face — aimed at making foreign-language releases feel native in each market instead of visibly overdubbed.

Where the line is

What separates these cases from the harms covered elsewhere on this page is consent and disclosure: the real person (or their estate) agreed to have their likeness or voice synthesized, and the use was disclosed rather than passed off as unedited footage of something that never happened. The same technology, used without that consent, is what the rest of this guide — and the deepfake laws explainer — is about.

Which deepfake guide do you actually need?

"Deepfake" now covers several very different problems, and this site has a dedicated guide for each one. Two recent industry reports show how fast the volume has grown: Gartner's September 2025 CISO survey found 62% of organizations had experienced a deepfake attack in the prior 12 months, and identity-verification firm Shufti's Identity Fraud Index projects deepfake identity fraud will rise roughly 495% in 2026 versus 2025. With that much volume across formats, the right next step depends on which format you're actually dealing with.

• Hearing a cloned voice, not seeing a fake video? Go to audio deepfakes for voice-scam tactics and the family "safe word" defense.
• Need to know if something is actually illegal, and what to file? Go to deepfake laws for the federal and state-by-state breakdown.
• Trying to tell a real deepfake from a cheaply edited clip (slowed-down video, a mislabeled photo)? Go to deepfake vs. cheapfake — cheapfakes are cruder but currently cause more real-world harm because they're so easy to make.

This page stays the general reference: what a deepfake is, how the technology works, and the harms and laws that apply across every format. Use the links above when you need the specifics for your situation.

FAQ: deepfake meaning, legality, and spotting

What is a deepfake in simple terms?

A deepfake is an AI-made image, video, or audio clip that makes it look like a real person did or said something they didn’t. It can be convincing enough to fool viewers, especially when shared without context.

Are deepfakes illegal in the U.S.?

Some are, some aren’t. As of spring 2026, 46 states have laws targeting AI-generated synthetic media (often political or sexual deepfakes). Federally, the TAKE IT DOWN Act (2025) criminalizes non-consensual intimate images (real or AI) and requires platforms to remove flagged content within 48 hours.

Is deepfake porn illegal?

Non-consensual intimate imagery—whether real or AI-generated—is targeted by the TAKE IT DOWN Act, signed May 19, 2025, with penalties up to 3 years in prison. Many states also have laws addressing sexual deepfakes, but the exact coverage varies.

How can I tell if a video is a deepfake?

Check for mismatched lighting, blurry hairline edges, unnatural blinking, and audio that doesn’t sync with mouth movements. When it matters, use tools like Microsoft’s Video Authenticator or Sensity AI as a second opinion.

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

Save evidence (screenshots, URLs), report it on the platform, and use relevant reporting channels. If a minor is involved, report to NCMEC. Adults can seek support through the Cyber Civil Rights Initiative. The TAKE IT DOWN Act also sets a 48-hour removal requirement after notice for flagged NCII.

Conclusion: what are deepfakes, and what you can do next

What are deepfakes? They’re AI-generated images, video, or audio that can convincingly depict a real person doing or saying something they never did—sometimes as a prank, but often as harassment, sexual abuse material, or fraud. With 46 states passing deepfake-related laws as of spring 2026 and new federal tools like the TAKE IT DOWN Act and the DEFIANCE Act, the legal system is starting to catch up—but your best protection is still fast documentation, smart reporting, and verification habits.

If you want to get more grounded in the wider pushback and accountability efforts around AI harms, explore /ai-backlash/ and practical options at /fighting-back/. For ongoing cases involving AI and platforms, see /ai-lawsuits/. And if you’re trying to understand adjacent impacts of the AI boom (from infrastructure to work), you can also check /data-center-map/ and /ai-layoffs/.

Not sure if a manipulated video is a deepfake or a cheapfake? Our best AI art detectors comparison covers the free and paid tools most useful for journalists and creators.

Deepfakes in elections and political campaigns

Political deepfakes are a distinct and fast-moving concern — a fabricated video or audio clip of a candidate saying or doing something they never did, timed to spread right before an election when there's no time to debunk it.

Regulators are starting to respond with labeling requirements rather than outright bans. The EU's AI Act, for example, includes a requirement that AI-generated or manipulated content resembling real people, places, or events be clearly labeled as artificially generated or manipulated. A growing number of individual countries and US states have layered on their own disclosure rules specifically for political ads and candidate depictions. Enforcement and exact requirements still vary widely by jurisdiction, so a labeling rule in one country or state doesn't necessarily apply elsewhere.