Deepfakes are synthetic or manipulated video, images, audio, or text that make people or events appear genuine when they are not. A convincing clip or voice note is not proof of identity or truth. Look for inconsistencies, check the source and context, then verify the person or request through a separate channel before sharing information, opening files, or sending money.
What is a deepfake?
A deepfake is synthetic media that has been generated or altered to falsely depict a person, voice, action, or event. It can include face swaps, digitally changed photographs, cloned voices, lip-synced video, fabricated livestreams, and computer-generated text. “Synthetic content” is the broader term used by the FBI; not every synthetic file is deceptive, but deepfakes are generally discussed in the context of misleading or harmful impersonation.
The FBI says content-creation tools have become broadly accessible and scalable. On September 12, 2023, CISA, NSA, and the FBI warned that synthetic-media threats were increasing and urged organizations to prepare to identify, defend against, and respond to them. The risk is not limited to celebrities: criminals use convincing voice, video, and image impersonation in payment fraud, account takeover, extortion, and social engineering.
How can you tell whether a video, image, or voice message is fake?
Use a sequence of checks rather than looking for one magical visual “tell.” Editing, compression, poor lighting, disability-related movement, or a bad connection can create the same artifacts as manipulation. A warning sign should trigger verification, not a definitive accusation.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
1. Check provenance and context first
- Identify who posted or sent the item, when it appeared, and whether the account or number is one the person normally uses.
- Look for confirmation on the purported speaker’s established website or official social channel, and check whether reputable outlets independently report the same event.
- Be suspicious of urgent demands, secrecy, unusual payment methods, requests for login codes, or instructions to bypass normal approval processes.
- Inspect the sender’s address, telephone number, URL, spelling, and contact details. A familiar name or profile picture is not authentication.
2. Inspect video and images
| Possible warning sign | What to examine | How to interpret it |
|---|---|---|
| Face or skin distortion | Blurred edges, changing facial geometry, odd skin color, or features that warp between frames | Useful for triage, but compression and low resolution can produce similar effects. |
| Eyes, eyebrows, hair, or blinking | Too much or too little blinking, eyebrows that do not move naturally, or hair that merges into the background | A cue to investigate further, not proof by itself. |
| Movement and positioning | Awkward head or body angles, unnatural motion, or hands and feet that change shape | Compare several frames and the person’s movement with the surrounding scene. |
| Lighting and shadows | Shadows pointing in inconsistent directions, reflections that do not match, or illumination that changes without a reason | Scene lighting can be difficult to judge in edited or heavily compressed footage. |
| Accessories and background | Glasses, jewelry, clothing details, signs, or background objects that deform or flicker | Small objects are often harder for generation systems to keep consistent. |
3. Listen to audio and lip synchronization
- Notice delays between speech and mouth movement, voice-call lag that does not fit the connection, or lips that do not form the sounds being heard.
- Listen for unnatural pauses, choppy sentences, unusual inflection, abrupt pitch changes, or phrasing the person would not normally use.
- Check whether background noise, room acoustics, and other speakers fit the claimed location. A clean studio-like voice over a noisy scene can be a mismatch.
The FBI’s 2025 impersonation alert also identified distorted hands or feet, unrealistic accessories, latency during calls, and unnatural movement as possible indicators. These clues can become less visible as generation tools improve, so visual or audio inspection should be treated as triage only. The FBI has warned that determining whether content was manipulated may become harder as synthetic media improves.
How do you verify a suspicious call, video, or message?
Verification must use a channel that the suspicious content did not control. Do not click a supplied link or call a number given in the message and treat the resulting conversation as confirmation.
Rank #2
- Pause the transaction. Do not send money, passwords, one-time codes, identity documents, confidential files, or cryptocurrency while the request is unverified.
- Find a trusted contact method yourself. Use a number from a saved contact, an official statement, a known company website that you typed yourself, or an established internal directory.
- Call or message through that independent route. Ask whether the person really sent the request and what they intended. For a financial or workplace request, use the normal approval process rather than an exception demanded in the message.
- Use a prearranged challenge. Families can use a secret word or question. Organizations can require a random phrase, a call-back to a known number, or a second authorized approver.
- Check the underlying event. Look for independent reporting, an official notice, or another primary source. A second account repeating the same fake is not independent confirmation.
- Preserve evidence. Save the original file or message, sender details, URL, call records, timestamps, and screenshots. Do not edit the original before making a copy.
What should you do if someone uses your face or voice?
Impersonation or financial fraud
- Tell the intended recipients through a verified channel that the material is fraudulent.
- Contact your bank, payment provider, employer, or account provider immediately if money, credentials, or access may be involved. Ask about recalls, freezes, password resets, and session revocation.
- Report the account, post, advertisement, or message to the platform using its impersonation or fraud process.
- Report suspected crime to the appropriate law-enforcement agency and retain the evidence package.
Intimate or abusive imagery
Image-based abuse can involve an authentic photograph, a digitally altered image, or an AI-generated intimate image of an adult or a minor. Use the platform’s reporting and removal process, preserve the original evidence, and follow the Federal Trade Commission’s current reporting route for digitally altered or AI-generated intimate images. If there are threats, extortion, stalking, or a child victim, contact law enforcement promptly rather than negotiating with the sender.
Do not redistribute the material “to prove” what happened. Share it only with the platform, investigators, legal counsel, or another service that specifically needs it, and follow that service’s instructions for secure submission.
Recommended Free Tools
Can deepfake detectors prove that something is fake?
No single detector should be treated as conclusive. Detectors can produce false positives and false negatives, and their performance varies with compression, language, editing, new generation methods, and the particular file. A detector result is one piece of evidence, not a substitute for identity verification or an investigation.
NIST’s November 20, 2024 synthetic-content transparency report describes a broader toolbox: provenance tracking, metadata, watermarking, labeling, synthetic-content detection, testing, auditing, and ongoing maintenance. Provenance can help show where a file came from and how it changed, but missing metadata does not prove fakery and metadata itself can be stripped or forged.
For remote identity proofing, NIST recommends analyzing known AI signatures, training reviewers to notice latency and synchronization problems, and adding random human-in-the-loop movement cues. For suspected face morphs, NIST’s August 18, 2025 implementation guidance emphasizes preventing morphs from entering operational workflows and combining automated analysis with human review and a documented investigation procedure.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How organizations should prepare for deepfake incidents
Organizations need controls that work even when a fake looks and sounds convincing. CISA, NSA, and FBI guidance frames preparation, identification, defense, and response as a lifecycle.
Best Value
Prevent and authenticate
- Use provenance or authenticity metadata and clear labeling when the platform supports them.
- Strengthen identity-proofing with liveness checks, random movement or response challenges, and a second factor that is independent of the media under review.
- Separate high-risk approvals: require a known call-back and another authorized person for unusual payments, payroll changes, password resets, or disclosure of sensitive data.
- Limit publicly available voice and video samples where doing so is consistent with the organization’s privacy and accessibility obligations.
Train people to recognize the attack pattern
- Teach staff that urgency, secrecy, authority, and emotional pressure are manipulation tactics, even when the voice or face appears familiar.
- Practice a simple rule: stop, use an independently sourced contact method, and document the result.
- Include audio, video, image, and identity-proofing examples in training; do not teach employees to rely on one artifact such as blinking.
Detect, review, and escalate
- Combine automated analysis with trained human review. Route uncertain or high-impact cases to a designated security, fraud, legal, or safeguarding team.
- Record the file, hash or original download where feasible, sender information, timestamps, detector outputs, reviewer decisions, and the independent verification performed.
- Define who can pause a payment, lock an account, notify affected people, contact a platform, and report to authorities.
- Test and maintain detection and provenance controls as generation techniques and platform formats change.
Balance security with privacy and accessibility
More checks can add cost, delay, data collection, and barriers for people with disabilities or limited connectivity. Choose controls proportionate to the consequence of being fooled, provide an accessible alternative verification route, restrict retention of biometric or voice data, and explain how evidence will be protected.
Is making or sharing a deepfake illegal?
There is no single rule that makes every deepfake lawful or unlawful everywhere. The answer depends on the content, intent, victim, harm, and jurisdiction. U.S. federal guidance and enforcement materials address conduct such as impersonation fraud, nonconsensual intimate imagery, extortion, and other crimes, but they do not establish one universal prohibition covering every synthetic image, parody, political communication, or fictional work.
Evaluate the specific act: who was depicted, whether consent was given, whether the material was intimate, whether it was used to obtain money or access, whether it was distributed, and where the people and services are located. Victims should preserve evidence and seek current advice from law enforcement, a qualified lawyer, or an appropriate victim-support service rather than relying on a general claim about “deepfakes.”
Quick Recap
A practical decision checklist
- Unexpected request? Stop and do not provide money, credentials, codes, or files.
- Could the media be manipulated? Check provenance, context, visual and audio inconsistencies, and urgency tactics.
- Need certainty? Verify using a trusted, independently found channel and a prearranged challenge.
- Possible harm or crime? Preserve originals, report to the platform, and contact the relevant authority.
- Organizational decision? Use layered provenance, human review, liveness or challenge checks, documented escalation, and periodic maintenance.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
