AI detection experts reveal how you can spot AI videos and stop scammers

Google's newest AI video generation model, VEO 3, is the most advanced AI video production tool yet, with the ability to make videos that are almost indistinguishable from real life. This provides many opportunities for creators, but it also gives bad actors the ability to create materials to use in scams and disinformation campaigns.

Christian Perry, CEO of AI detector Undetectable AI, is an expert in spotting the hallmarks of generated content and has highlighted the best ways to identify potentially harmful generated content.

Uncanny faces

"Facial expressions are still a weak spot for AI video generation," commented Perry. "AI has a hard time producing natural movements consistently, with areas that have lots of small details, such as faces, providing tell-tale signs that a video could be fake.

"Sometimes, AI-generated characters have expressions that don't convey the same emotions as their speech or body language. For example, a character might be speaking calmly, but with an angry face."

"However, generated faces aren't always so easy to spot. You often have to look closely at how a face moves and how each part links with the rest. In isolation, a movement might look natural, but when combined with another element it looks uncanny, like a mouth that opens too wide or facial muscles not matching what the character is doing with their face.

"This is most pronounced in the eyes. AI-generated eyes can lack details like defined whites (sclerae), irises, and pupils. Generated videos frequently have a character's eyes dart from one focus point to another without a good explanation. Characters will also have their eyes move to the extremes of their range of motion, and can look like they are looking at two things at once.

"Blinking often looks unnatural in generated videos, too. Common errors include: eyelids not closing fully, blinking that looks forced rather than a natural reflex, and blinks at consistent intervals."

Background details

Background elements feature logical inconsistencies more often than foreground elements, as details that would be portrayed the same by real people and objects no matter where they are in a scene, become exponentially more difficult for AI systems to depict correctly in the background.

Perry expanded on this: "Pay attention to what's going on in the video's background, as the logic of a scene usually breaks down in these areas first. An AI only has so much processing power, and this means the foreground elements are prioritized, with background elements usually left in a state to appear good enough without major scrutiny."

Mismatched audio

Audio is another key way of identifying AI content. AI-generated speech may sound convincingly human, but it frequently lacks the depth and nuance that a real voice has. Real recordings will have slight quirks such as breaths and vocalizations, while AI-generated audio will either lack these or will insert them in unnatural places.

"AI still has trouble matching the right audio to what's happening on screen," said Perry. "You often find unnatural speech patterns from people, and ambient noises that don't match the scene. For example, rolling waves in the middle of an ocean might sound like gentle splashes in a pool."

How to protect against scams

According to the FBI, criminals now use AI video technology to create fake people during video calls or imitate people the victim might know to support their fraudulent schemes. This generated content will only become more convincing as more powerful tools such as VEO 3 reach the public.

Perry commented on how AI can be used in scams and disinformation campaigns, and how you can protect yourself: "Scammers use AI technology to create fake content to mislead or elicit an emotional response. This content is then used to convince their victim to part with money or valuable information.

"The best way to protect yourself if you think you're being targeted by an AI scam is to verify what you're seeing with trusted sources. There's no substitute for checking things with multiple news sources or people you trust, so always take the time to make sure what you're seeing is genuine.

"Scammers don't usually wait around, since they know that the longer you take to check things, the less likely you are to fall victim to their schemes. If something is genuine, you'll almost certainly have the time to do your due diligence to check that it is actually the truth."

This information was provided by undetectable.ai