Autoscaling is focused on meeting capacity targets, not preserving individual instances.

If scale-in policies are aggressive and instance protection isn’t enabled, the autoscaler will happily terminate healthy instances to reduce capacity. From its perspective, everything is working as designed.

Problems arise when workloads aren’t prepared for termination or don’t drain gracefully before shutdown.

Takeaway: Autoscaling protects numbers, not workloads, unless you configure it to.