AI Safety – What is a Deepfake?

A deepfake is a type of manipulated video, image, or audio created using artificial intelligence that makes it look or sound like someone did or said something they never actually did. The concern is that deepfakes can be so realistic that they can be used to spread false information, scams, or impersonations.

How does it work?

A hidden layer of artificial neurons between the input and output layers of a neutral network to convert a real image into a fake one. The process of creating a deepfake involves two algorithms. The “generator” is to create the best replica the “discriminator” is to detect the deepfake, so that the “generator” can correct its inconsistencies. 

Deepfakes can be used for fun, education or science, but sometimes they are used for political misinformation, cyberbullying, nudifiers or scams. Most deepfakes look unrealistic and can be spotted immediately, but sometimes they can be extremely convincing. 

How to spot deepfake videos/images:

While AI is getting better at generating photos and videos, most deepfakes still have subtle flaws. Here’s what to look for:

 Facial & body oddities: 

  • Watch for unnatural blinking, distorted ears, smudgy chins, or inconsistent skin texture—look closely at eyes, mouth, fingers, and body alignment
  • AI often misaligns speech and mouth movement—especially with sounds like “p,” “b,” “m”

 Lighting & shadow mismatches: 

  • Real shadows behave predictably. If shadows do not match the environment or there is inconsistent lighting across the face, it might be AI generated.

Texture/edges Imperfections: 

  • Some deepfake images have too-smooth skin, weird pixel noise patterns or fuzzy outlines around the head or edges. 

Audio glitches: 

  • Voice clips may contain robotic artifacts or vocoder traces
  • Researchers have built models to detect these, such as Spectrogram based models or Recurrent Neural Networks

Source verification:

  •  Always cross-check with trustworthy news outlets or use reverse-image searches to find originals, like Google Reverse Image Search or InVID to see if the content has appeared before another content. 

AI based detectors:

  • Find tools that can analyze videos or audios for inconsistencies

Easy-to-Use Tools To Help You Detect Deepfakes: 

  • Facia AI: fast at detecting liveness in videos. This is the only free one.
  • Vastav AI: Real-time detector for images, videos, and voice using forensic metadata analysis.
  • Deepfake detector: AI tool that can identify if an audio or video is a deepfake or real.
  • Deepware Scanner: Scans uploaded videos/images and flags AI manipulation.

Stay Safe in an AI-Generated World

Real-World Risk: The Deepfake Kidnapping Scam 

Imagine receiving a video or phone call that looks and sounds just like your sibling, friends or family begging for help. That is the terrifying reality of deepfake scams. There have been alarming, real cases where AI-generated videos falsely showed someone’s kidnapping, fooling families and causing panic. 

Experts warn that these scenarios are becoming more common: attackers can target individuals with deepfakes tailored to personal data, fabricating threats from loved ones. Having a “safe word” to use among family members is a strong defense. A Deepfake scammer or AI impersonator wouldn’t know this secret word, because they only have your public information, but if someone was physically with your loved one, then they could potentially force the victim to reveal the word. 

AI Nudifiers & Cyberbullying 

Deepfakes are not just used to fake news – they are also contributing to harassment and abuse. AI nudifiers can create nonconsensual nude images— “in fact, up to 96% of deepfake videos are pornographic” (Deeptrace, 2019, as cited in MIT Sloan, 2020). Students have used deepfakes to falsely accuse peers or teachers, triggering humiliation, depression, or social withdrawal. AI-generated harassment is scarier than traditional bullying, as messages can be hyper-specific and emotionally harmful even if they are fake . 

Preventive Actions: 

Document incidents (screenshots, timestamps). 

  • Making a screenshot, saving a message, and information about who shared them. Platforms and authorities almost always require concrete evidence when investigating online bullying and in order to take  action against the offender. 

 Report:

  • To school or a trusted adult; reach out to helplines like 988 Crisis Lifeline if needed.
  • To social media platforms where nudifiers and cyberbullying were spread, so it can be removed.
  • To police, to the National Center for Missing & Exploited Children (NCMEC) if the nudifier image is a minor.  

Practice:

  • The better you are at detecting deepfakes, the safer you’ll be.
  • Take online quizzes to practice detecting deepfakes, like this one from Northwestern University.

Be Safe:

  • Update privacy settings
  • Avoid sharing personal photos or locations
  • Learn about deepfake technology
Scroll to Top