Autoscaler will respond differently to different traffic patterns, but won't go as far as identifying a particular traffic pattern as malicious and refusing to

Autoscaler will respond differently to different traffic patterns, but won't go as far as identifying a particular traffic pattern as malicious and refusing to grow resources to accommodate it (most architectures filter malicious requests upstream of the instance itself, eg. at the load balancer or via a proxy). Even if you could identify malicious usage, it's still a matter of policy on how you want to respond to it - given that if it can reach your VM it will likely be affecting legitimate traffic too.