China Warns Against Immediate Speech on Unknown Calls Over AI Voice Cloning Risks
China's Ministry of State Security advised citizens to pause for three seconds when answering phone calls from unknown numbers to avoid having their voice sampled. The security agency warned that scammers deliberately stay silent after connecting to prompt victims into speaking, capturing just a few seconds of audio to build AI voice clones for imposter fraud. Modern generative AI tools can clone an individual's vocal characteristics using minimal audio samples, significantly lowering the barrier for convincing deepfake fraud. The advisory underscores a broader shift where personal biometrics—such as voiceprints, facial patterns, and gait—are transformed into harvestable data that threaten individual privacy and national security. The Ministry noted that AI systems can combine stolen biometrics like facial images and gait data to infer physical traits, emotional states, or family relationships for targeted exploitation. Authorities emphasized a 'provide only when necessary' principle for biological data and reminded the public that unauthorized collection or sale of personal biometrics carries criminal liability.
## BACKGROUND
Voiceprint recognition and voice cloning rely on capturing unique acoustic features of a speaker, including vocal tract geometry and pitch nuances. Recent advances in zero-shot text-to-speech (TTS) models allow neural networks to recreate a voice profile with high fidelity using only seconds of reference audio. Similarly, gait recognition identifies individuals by analyzing their walking patterns and body posture from video footage.