This is statistics though. Edge cases are nothing new and risk management concepts have evolved around fat tails and anomalies for decades. Therefore the statem

This is statistics though. Edge cases are nothing new and risk management concepts have evolved around fat tails and anomalies for decades. Therefore the statement is as naive as writing a trading agent that is 100% correct. In my opinion, this error shows lack of understanding responsible scaling architectures. If this would be their first screw up I wouldn't mind, but Google just showed us a group of diverse Nazis. If there is a need for consumer protection for online services, it is exactly stuff like this. ISO 42001 lays out in great detail that AI systems need to be tested before they are rolled out to the public. The lack of understanding of AI risk management is apparent.