Recognizing and Responding to Deepfakes and Voice Cloning

AI-generated images, audio, and video — broadly called synthetic media or “deepfakes” — are no longer the exclusive domain of sophisticated actors. Tools available to the general public can clone a voice from a brief recording, place someone’s face on another person’s body, or generate a convincing photorealistic image from a text description. The cost in time and technical skill has dropped to near zero.

The practical consequence is that fabricated media can now be produced by abusive partners, adversaries in legal disputes, or bad actors online. Images that appear to show you doing or saying something you did not do. A phone call that sounds like a family member in distress. Evidence submitted in a legal proceeding that was never captured by a camera.

This tutorial covers how to recognize synthetic media, how to verify authenticity before acting on it, and what to do when fabricated content is used against you. The steps apply whether you are the subject of the fabrication or whether you have received content you suspect may be artificial.

Steps

1. Understand what synthetic media can and cannot do — yet. Current AI image generation consistently struggles with specific areas: hands (extra or fused fingers, unnatural joint angles), teeth (irregular number, texture, spacing), background edges (blurring or smearing where a subject meets the background), and bilateral symmetry (earrings that don’t match, uneven facial features). In video, look for unnatural blinking, skin that appears too smooth or slightly misaligned with the skull, and mouth movements that don’t fully sync with audio. These artifacts are not always present, but their absence does not confirm authenticity.

2. Run a reverse image search before acting on an image. Right-click any image and choose “Search Image” (Chrome/Edge) or use Google Images, TinEye, or Bing Visual Search. A reverse search checks whether the image exists elsewhere on the web and can surface the original source. If an image of you appears in a context that seems impossible or falsified, a reverse search may reveal whether it was lifted from another context or composed from identifiable source images. Note: an image that doesn’t appear in any reverse search results is not confirmed as AI-generated — newly created images often have no indexed history.

3. Check image metadata. Every digital image contains EXIF metadata: camera model, GPS coordinates, date and time, software used to process it. Right-click → “Get Info” on Mac or use a free tool like Jeffrey’s Exif Viewer to inspect the file. AI-generated images typically have no EXIF data, or metadata that shows image-editing software rather than a camera. A photo claiming to show a specific place and time should have GPS and timestamp data consistent with that claim. Absence or inconsistency is a red flag — not proof — as metadata can be stripped or spoofed.

4. Use AI detection tools as a secondary check. Several tools attempt to classify whether an image or audio file was AI-generated: Google’s SynthID (for Google-generated content), Hive Moderation, and Illuminarty for images; AI or Not for audio and images; FakeCatcher for video. These tools have meaningful false-positive and false-negative rates — treat their output as one signal among several, not a verdict. A “likely AI-generated” result is a reason to investigate further, not a conclusion to present as fact.

5. Verify voice identity through a second channel before acting. Voice cloning tools can produce convincing audio from under a minute of recorded speech. If you receive an unexpected call from a known person making an urgent or unusual request — a family member claiming an emergency, a lawyer asking you to act immediately, an authority figure making a demand — do not comply based on voice recognition alone. Hang up and call back on a number you already have for that person, or contact someone who can physically verify their location. Establish a code word with close contacts for emergency calls that only genuine family members would know.

6. Preserve all evidence before reporting to platforms. Platform reporting processes frequently result in content removal — which is the goal, but not before you’ve documented what you found. Before filing any report: screenshot the content including the account name, URL, and any visible timestamps; download the file if the platform permits; record the URL and note the date and time you accessed it; use a screen recording if the content is video or audio. Once documented, report through the platform’s non-consensual intimate imagery or synthetic media policies if applicable. Major platforms have dedicated reporting pathways for fabricated content used for harassment.

7. Report fabricated content as evidence, not only as a terms-of-service violation. If synthetic media is being used to harass, coerce, or blackmail you — or to introduce false evidence into a legal proceeding — contact law enforcement and document your report. Several U.S. states have laws specifically criminalizing non-consensual deepfakes and AI-generated intimate imagery. Keep the case number from any law enforcement report — it can be used to compel faster platform response and is necessary for civil legal action. Attorneys who specialize in online harassment or cyberlaw can advise on state-specific options.

8. Build a proactive record of your authentic media. If you have reason to believe you may be targeted, periodically date and archive authentic photos, videos, and voice recordings in encrypted storage under your sole control. A timestamped archive of genuine media makes it easier to demonstrate temporal inconsistencies in fabrications and supports claims that you were somewhere other than where a fabricated image places you.

Common Mistakes

• Deleting content before preserving it. Content you report may be removed quickly. Always document before you report. An undocumented claim of fabricated media is significantly harder to act on legally or formally.
• Trusting voice alone for identity verification. Voice has always been a weak authentication factor; voice cloning makes it weaker. Any request delivered by phone that is unusual, urgent, or asks you to take an irreversible action should be verified through a separate channel before you comply.
• Assuming AI-generated content is always visually obvious. Artifact-spotting and detection tools are useful, but current AI generation produces output that passes casual inspection. The goal is not to catch every fake — it is to not act on unverified content when the stakes are high.
• Treating AI detection tool output as authoritative. These tools are probabilistic. Neither a “real” nor “AI-generated” result is sufficient on its own for legal proceedings or formal complaints. Use them as investigative leads that warrant further verification.
• Assuming platforms will act quickly. Platform review timelines vary and are not guaranteed. Law enforcement reports, attorney demand letters, and documented escalation typically produce faster results than standard user reports in serious cases.

Resources

• Documenting Interactions with Authorities — how to preserve evidence and maintain a usable record
• Setting Up Encrypted Communication on Apple Devices — protecting your communications from interception
• AI Surveillance and Reducing Your Exposure — how AI tracking and profiling work
• Survival Resources: Legal — legal aid, cyberstalking resources, and support organizations
• Glossary — definitions for technical terms used in this tutorial