While AI technology is advancing rapidly, human verification remains the most effective defense against deepfakes. Because scammers often combine high-quality AI generation with high-pressure tactics, the goal is not just to detect the technology, but to verify the identity and intent of the sender.
Here are practical methods users can employ to spot potential AI manipulation:
1. Check for Visual Inconsistencies (Video/Images)
Current AI video generation often struggles with fine details and physics. Watch for:
- Lip Syncing Issues: Does the person’s mouth movement match the audio exactly? AI sometimes produces a “slurred” or lagging lip movement.
- The “Uncanny” Look: Skin may look unusually smooth, waxy, or plastic-like. Look for weird lighting or shadows that don’t match the environment.
- Eyes and Blinking: AI-generated faces sometimes have unnatural blinking patterns (blinking too rarely or too often) or pupils that do not react to light.
- Background and Hands: Backgrounds may blur, morph, or contain nonsensical objects. AI often struggles with hands (extra fingers, missing fingers) and accessories (jewelry that looks melted or glitchy).
- Teeth and Tongue: Teeth may look like a block of white or move unnaturally.
2. Listen for Audio Anomalies (Voice Messages)
AI voice cloning can be convincing, but listen for subtle unnatural qualities:
- Lack of Breathing: Human speech naturally includes inhales and exhales. AI audio sometimes lacks these or includes robotic “clicks” between sentences.
- Rhythmic Uniformity: AI speech often has a perfectly consistent cadence, lacking the natural pauses, stutters, or speed changes of a stressed or excited human.
- Background Noise: If the voice sounds crystal clear (studio quality) but the background sounds like a normal room or street, the voice may be layered over a video.
- Emotional Dissonance: Does the tone match the urgency of the message? If they are claiming an emergency, they might sound oddly calm or flat.
3. The “Challenge” Test
If you receive a video or voice message from someone you know requesting money or help, do not trust it immediately. Challenge the sender in real-time:
- Ask for a Specific Movement: In a live video call, ask them to cover their mouth with their hand, turn their head 360 degrees, or blink three times quickly. AI deepfakes often struggle with complex, real-time commands.
- Ask a Personal Question: Ask a question that requires specific knowledge you two share that isn’t publicly available (e.g., “What was the name of the cat we had in 2015?”).
- Request a Live Call: If they are sending pre-recorded videos, insist on a live call. Scammers often avoid live interaction where they cannot control the narrative.
4. The “Golden Rule”: Independent Verification
This is the single most effective method, regardless of whether AI is used:
- Contact Them Separately: Put the device away and call the person on a known, trusted phone number or use a different platform you already use with them. Do not use the contact information provided in the suspicious message.
- Ask Directly: Say, “Did you just send me that video asking for help?” If they deny it or claim they haven’t heard from you, the message was a scam.
- Verify the Account: Check the social media account that sent the message. When was it created? Does it have a history of posts? If the account is new or the activity doesn’t match the person’s usual behavior, it is likely compromised or fake.
5. Technical Checks (Limited Utility)
- Reverse Image Search: If the profile picture or video thumbnail is AI-generated or stolen, you might find the original source elsewhere on the internet using Google Lens or TinEye.
- Metadata Analysis: Check the file properties. Sometimes AI-generated images or videos have different file structures than standard camera outputs, though savvy scammers often edit this metadata to look normal.
Important Disclaimer
Do not rely solely on detection. AI technology is improving faster than detection tools. A scammer can use a high-quality video that passes visual checks but is still fraudulent. Always assume the message is fake until you have verified it through a separate, secure channel. If the message creates a sense of urgency (“Do this now or else”), it is almost certainly a scam.