I just saw that the latest iphone update allows you to unlock the phone with a facemask on, which got me thinking: I'm surprised there haven't been more published exploits that make use of this attack vector. While Apple might have the means to make this relatively robust, I'd suspect the fractured android ecosystem would lead to a lot of half-baked implementations, and the literature on the fragility of ML models w/r/t adversarial input would suggest that even Apple's model should still be susceptible. I recall there being a lot of hubub over police having the tools to crack iOS encryption, which would seem like a harder task than fooling a facial recognition algorithm. How are there so few exploits for facial recognition login floating around?