Deepfake
Deepfake is a digital media synthesis method based on deep learning, using algorithms such as Generative Adversarial Networks (GANs) to perform highly realistic replacement or fabrication of a person's facial appearance, voice, or body movements, producing audio-visual content that is difficult to distinguish from authentic material with the naked eye. The technology has legitimate applications in entertainment and artistic creation, while also raising broad social and ethical concerns regarding information security, privacy protection, and manipulation of public opinion.
Core Facts
Timeline (last 90 days)
The social harm of deepfakes lies in causing people to no longer believe any photo or video, dismantling the mechanism of trust.
Countries are exploring dedicated legislation on deepfakes, but the speed of legal updates lags far behind the iteration of AI generation capabilities, and the cross-border nature of dissemination limits the effectiveness of regulation by any single country.
Practical principles for ordinary people to identify deepfakes include: check the source, watch for "too perfect" narratives, do a reverse search, pay attention to platform labels, and delay sharing.
Deepfake models need only dozens to hundreds of photos of a target person to learn their facial features and synthesize new scenes.
The reasons celebrities become a deepfake hotspot include: extremely abundant source material, easily manipulated public emotions, and传播 costs approaching zero.
AI-generated misinformation (including deepfake images, AI-mass-produced inflammatory copy, and bot accounts disguised as real users) makes rumor identification harder, and traditional keyword filtering and manual review struggle to keep up with its generation speed.
Many model providers choose to directly intercept the generation and editing of videos involving real individuals at the API level to avoid legal and ethical risks such as portrait rights violations and deepfake misuse.
As deepfakes and generative AI become more prevalent, disputes over distinguishing authentic records from AI re-creations will become increasingly common.
Some U.S. states have made it illegal to create deepfake imagery of others without their consent.
California expanded relevant privacy law provisions targeting deepfake content in 2023.
16 more timeline events
All Facts (20)
Diffusion-model-based face swapping and driving technology requires only a small number of reference images to transfer a target person's facial features onto arbitrary video footage.
65%UnverifiedCountries are exploring dedicated legislation on deepfakes, but the speed of legal updates lags far behind the iteration of AI generation capabilities, and the cross-border nature of dissemination limits the effectiveness of regulation by any single country.
50%UnverifiedPractical principles for ordinary people to identify deepfakes include: check the source, watch for "too perfect" narratives, do a reverse search, pay attention to platform labels, and delay sharing.
50%UnverifiedThe social harm of deepfakes lies in causing people to no longer believe any photo or video, dismantling the mechanism of trust.
50%UnverifiedDeepfake models need only dozens to hundreds of photos of a target person to learn their facial features and synthesize new scenes.
50%UnverifiedThe reasons celebrities become a deepfake hotspot include: extremely abundant source material, easily manipulated public emotions, and传播 costs approaching zero.
50%UnverifiedAI-generated misinformation (including deepfake images, AI-mass-produced inflammatory copy, and bot accounts disguised as real users) makes rumor identification harder, and traditional keyword filtering and manual review struggle to keep up with its generation speed.
50%UnverifiedMany model providers choose to directly intercept the generation and editing of videos involving real individuals at the API level to avoid legal and ethical risks such as portrait rights violations and deepfake misuse.
50%UnverifiedSome U.S. states have made it illegal to create deepfake imagery of others without their consent.
50%UnverifiedCalifornia expanded relevant privacy law provisions targeting deepfake content in 2023.
50%UnverifiedModulate is deploying its models to detect deepfakes, fraud, and scam activity.
50%UnverifiedModern generative AI can reconstruct highly realistic audiovisual likenesses from just a few photos and tens of seconds of audio.
50%UnverifiedSome U.S. states have passed bans targeting deepfakes related to elections.
50%UnverifiedDeepfake technology poses a challenge to the authenticity of information
50%UnverifiedThere are mainly two directions for addressing the risks of AI-generated content: content provenance technology and AI-generated content detection models, but both approaches currently have significant limitations.
50%UnverifiedDeepfakes have already caused real harm in scenarios such as political opinion manipulation, financial fraud, and non-consensual pornographic content.
50%UnverifiedDeepfakes refer to using generative models to replace the face or voice of a real person in another video to create convincingly deceptive content.
50%UnverifiedVideo and image models face challenges that have not yet been fully resolved, including copyright and training data disputes, content authenticity and deepfake risks, long-video coherence bottlenecks, and generation costs.
50%UnverifiedWith the help of AI image generation and deepfake technology, the cost of creating celebrity sickbed photos or tribute statement videos has dropped significantly.
50%UnverifiedA consumer-grade computer combined with open-source models can generate deepfake content capable of deceiving most people.
50%