How to Spot a Deepfake on Social Media | McAfee Blog

Think you can spot a fake on social media? It’s getting tougher. Particularly as deepfake technology gets far better and far easier to use.

Here’s why that matters.

You might find yourself among the 50% of Americans who say they get their news on social media at least “sometimes.”[i] Plenty of deepfakes deliberately pose as legitimate news. You might also stumble across promos or deals on social media. Scammers create yet more deepfakes for phony giveaways and bogus investment opportunities.

In short, what you’re seeing might be a fake. And your odds of stumbling across a deepfake on social media are on the climb.

That means using social media today requires more scrutiny and skepticism, which are two of your best tools for spotting deepfakes.

The best way to spot deepfakes right now

Whether you’re staring down AI-generated text, photography, audio, or video, some straightforward steps can help you spot a fake. Even as AI tools create increasingly convincing deepfakes, a consistent truth applies — they’re lies. And you have ways of calling out a liar.

Slow down.

Malicious deepfakes share something in common. They play on emotions. And they play to biases as well. By stirring up excitement about a “guaranteed” investment or outrage at the apparent words of a politician or public figure, deepfakes cloud judgment. That’s by design. It makes deepfakes more difficult to spot because people want to believe them on some level.

With that, slow down. Especially if you see something that riles you up. This offers one of the best ways to spot a fake. From there, the next step is to validate what you’ve seen or heard.

Consider who did the posting.

Because what you’re seeing got posted on social media, you can see who posted the piece of content in question. If it’s a friend, did they repost it? Who was the original poster? Could it be a bot or a bogus account? How long has the account been active? What kind of other posts have popped up on it? If an organization posted it, look it up online. Does it seem reputable? This bit of detective work might not provide a definitive answer, but it can let you know if something seems fishy.

Seek another source.

Whether they aim to spread disinformation, commit fraud, or rile up emotions, malicious deepfakes try to pass themselves off as legitimate. Consider a video clip that looks like it got recorded at a press conference. The figure behind the podium says some outrageous things. Did that really happen? Consult other established and respected sources. If they’re not reporting on it, you’re likely dealing with a deepfake.

Moreover, they might report that what you’re looking at is a deepfake that’s making the rounds on the internet. Consider the Taylor Swift “Le Creuset scam” of early 2024. News outlets quickly revealed that the singer was not giving away free, high-end cookware.

A technique called SIFT can help root out a fake. It stands for: Stop, Investigate the source, Find better coverage, and Trace the media to the original context. With the SIFT method, you can indeed slow down and determine what’s real.

Have a professional fact-checker do the work for you.

De-bunking fake news takes time and effort. Often a bit of digging and research too. Professional fact-checkers at news and media organizations do this work daily. Posted for all to see, they provide a quick way to get your answers. Some fact-checking groups include:

What are typical signs of a deepfake?

This gets to the tricky bit. The AI tools for creating deepfakes continually improve. It’s getting tougher and yet tougher still to spot the signs of a deepfake. The advice we give here now might not broadly apply later. Still, bad actors still use older and less sophisticated tools. As such, they can leave signs.

How to spot AI-generated text.

Look for typos. If you spot some, a human likely did the writing. AI generally writes clean text when it comes to spelling and grammar.

Look for repetition. AI chatbots get trained on volumes and volumes of text. As such, they often latch onto pet terms and phrases that they learned as they were trained. Stylistically, AI chatbots often overlook that repetition.

Look for style (or lack thereof). Today’s chatbots are no Ernest Hemingway, Mark Twain, or Vladimir Nabokov. They lack style. The text they generate often feels canned and flat. Moreover, they tend to spit out statements, yet with little consideration for how they flow together.

How to spot deepfake photos.

Zoom in. A close look at deepfake photos often reveals inconsistencies and flat-out oddities. Consider this viral picture of the “Puffer Pope” that circulated recently. Several things point toward a bogus image.

Credit:CNNStart with the hands in the image. The right hand isn’t fully formed. Many AI tools have a notoriously tough time with rendering fingers properly. Meanwhile, the left hand features some lighting and skin tones that look a bit unnatural. An even closer look shows that the crucifix worn by the Pope only has half a chain. Next, look at the face and the unusual shadows cast by the glasses he wears.

How to spot deepfake audio and video.

Keep an eye on the speaker. A close look at who’s doing the talking in a deepfake video can reveal if it’s a fake. Subtle things reveal themselves. Is the speaker blinking too much? Too little? At all? How about their speech. Does it sync up with their mouth perfectly? These might be signs of a deepfake.

Watch how the speaker moves. In the example of the Ukrainian presidential deepfake, it appears that only President Zelensky’s head moves. Just slightly. This is a sign of lower-grade video deepfake technology. It has difficulty tracking movement. Another possible sign is if the speaker never moves their hand across their face. Once again, that might indicate the work of lesser AI tools. In that case, they render the facial image on the hand.

Look at and listen to the context. If a speaker is in an open public space, does it sound like they’re speaking in that environment? For example, if they’re in a city park, can you hear birds? What about traffic noise? How about the murmurs of the crowd? If that’s missing, or it feels like ambient sounds are piped in like the laugh track in an old sitcom, you might have a deepfake on your hands.

How does the speaker sound? In the case of audio-only deepfakes, today’s AI tools work best when they’re fed smaller chunks of text to create speech. They don’t work as well with big blocks. This requires creators to stitch those chunks together. As a result, the cadence and flow might sound on the copy side. Also, you might not hear the speaker taking breaths, as normal speakers do.

Be skeptical. Always.

With AI tools improving so quickly, we can no longer take things at face value. Malicious deepfakes look to deceive, defraud, and disinform. And the people who create them hope you’ll consume their content in one, unthinking gulp. Scrutiny is key today. Fact-checking is a must, particularly as deepfakes look sharper and sharper as the technology evolves.

Plenty of deepfakes can lure you into sketchy corners of the internet. Places where malware and phishing sites take root. Consider using comprehensive online protection software with McAfee+ to keep safe. In addition to several features that protect your devices, privacy, and identity, they can warn you of unsafe sites too. While it might not sniff out AI content (yet), it offers strong protection against bad actors who might use fake news to steal your info or harm your data and devices.

[i] https://www.pewresearch.org/journalism/fact-sheet/social-media-and-news-fact-sheet