Can liveness detection models generalise to synthetic media generation techniques they were never trained on? [D]
Current liveness detection systems may struggle to generalize to new synthetic media generation techniques they were not trained on. Models trained on older deepfake samples might not be effective against emerging generation methods. This raises questions about the update cycles for vendors offering deepfake detection capabilities, especially when they provide confident but unspecific answers regarding this temporal gap. AI
IMPACT Questions the effectiveness of current AI models in detecting novel synthetic media, potentially impacting security and identity verification systems.