Remote identity verification assumes that the camera stream shows a live person. But public real-time face-swap tools can replace that person before the stream ever reaches the verifier, undermining that assumption. A new paper on arXiv examines this threat specifically for interactive liveness detection, where a user is prompted to perform actions or respond to challenges.
The authors argue that not all challenges are equal: some are far more vulnerable to real-time face-swapping than others. They evaluate interactive liveness detection against these attacks and contrast it with passive deepfake detectors, which they say generalize poorly to the real-time scenario. The abstract is truncated, so full details on methodology and results are not yet available, but the core claim is clear: verification systems need to consider which specific challenges they use, not just whether they include liveness detection at all.