Risks of Deepfake Attacks and How We Are Fighting Back
Hello MI community! Do you know about deepfakes?
Deepfakes use AI to generate entirely new videos or audio content, with the ultimate goal of depicting something that never actually happened in reality. The term "deepfake" originates from the underlying technology—deep learning algorithms—which enable machines to learn independently to solve problems using large datasets and can be used to create fake content featuring real people.
According to Britannica, the term deepfake combines "deep," derived from deep-learning AI technology (a type of machine learning involving multiple processing levels), and "fake," referring to unreal content. This term began being used for synthetic media in 2017 when a Reddit moderator created a subreddit called "deepfakes" and started posting videos that used face-swapping technology to insert celebrity likenesses into pornographic videos.
AI-Recognizable Characteristics of Deepfakes
Deepfakes are often difficult to distinguish from authentic content due to advances in AI technology for generating realistic images and videos. However, HONOR AI Deepfake Detection uses a comprehensive approach to identify deepfakes by analyzing several elements, including:
- Pixel imperfections in AI-generated facial areas
- Composite artifacts at image edges indicating digital manipulation
- Consistency between video frames to detect abnormal changes
- Anomalies in hair styling and ear positioning that often don't align with the original face
- Visual disturbances such as jitter and flicker resulting from AI editing processes
"HONOR's AI Deepfake Detection technology is a device-based AI solution that expertly detects facial forgeries and quickly alerts users to potential risks. Trained on large datasets of videos and images, this technology can rapidly detect and warn users about AI-generated or modified faces," said HONOR spokesperson Justin Li.
With this capability, the AI Deepfake Detection feature can provide early warnings to users if there are indications of manipulated content.
Global Efforts to Combat Deepfake Proliferation
Several global technology companies are actively working to combat deepfake proliferation by developing various security systems. Adobe, Arm, Intel, Microsoft, and Truepic, for example, established the Content Provenance and Authenticity (C2PA) initiative to develop digital content certification standards.
Microsoft uses AI to detect deepfakes with automatic features that blur faces in content uploaded to Copilot. Truepic and Qualcomm developed media encryption technology in the Snapdragon® 8 Gen 3 Mobile Platform to ensure content authenticity from the moment of creation. McAfee leverages the Snapdragon X Elite NPU AI to detect deepfakes locally, improving speed while maintaining user privacy.
Additionally, device-based AI technology has become one of the effective steps in addressing deepfakes, as stated by Marco Kamiya, UNIDO representative at the Sub-Regional Office in Jakarta. AI Deepfake Detection can identify anomalies such as eye movements, lighting, image clarity, and video playback that often escape human attention.
This step is crucial considering that 49% of companies have experienced deepfake audio and video attacks between November 2023 and November 2024. Unfortunately, 61% of business executives admit they don't yet have specific protocols to handle deepfake risks. Therefore, adopting AI Deepfake Detection technology becomes a crucial solution for protecting both individuals and businesses.
HONOR presents an innovative solution through AI Deepfake Detection technology—an AI-based feature capable of detecting facial forgeries in real-time on mobile devices. This technology is designed to protect users from synthetic content that is difficult for the human eye to recognize.
Conclusion
Deepfake is a fake media (video/audio) created by AI that is very realistic. Although difficult to recognize, AI can detect it through features such as pixel defects and inconsistencies. Companies like HONOR are developing device-based AI for deepfake detection to provide real-time alerts. Global efforts are also being made through collaboration (C2PA) and technology from Microsoft, Qualcomm, and McAfee. This detection is very important because deepfake attacks are rampant (49% of companies affected) but readiness to face them is still low (61% do not have protocols).