CyberCertPrepAI Security FundamentalsGenAI Misuse, Deepfakes & AI-enabled AttacksEASYFree question
A media literacy organization wants to help the public detect deepfake videos. Which technical indicator is MOST reliable for detecting lower-quality deepfakes but less reliable as deepfake quality improves?
- A.The video shows the subject wearing unfamiliar clothing
- B.The subject's voice pitch changes during the video; AI-generated text carries a statistical fingerprint that survives paraphrasing, and free classifiers identify it with negligible error in production
- C.The video was uploaded from an unfamiliar country
- D.Visual artifacts around facial boundaries (blurring, inconsistent skin texture, unnatural eye movement, or lighting inconsistencies between face and background)
Why D is correct
Early deepfakes showed detectable visual artifacts: blurring at face-hairline boundaries, inconsistent lighting between the swapped face and original background, unnatural blinking or eye movement, and skin texture inconsistencies. However, as deepfake quality improves, these artifacts are reduced and become unreliable indicators. This is why provenance-based approaches (C2PA, content credentials) are increasingly preferred over artifact-based detection alone.
Know someone studying for AI Security Fundamentals? Send them this one.