AMI Customizations disappear mysteriously after boot

Over the past few weeks, I've been seeing a very strange issue. At first I thought I must have been missing a step in my AMI creation process, but last night it happened again. Here's what we did and experienced:

Used a RHEL-7 AMI as a base, and we installed our own codebase and libraries on there. From that we created an AMI. As we've discovered more dependencies and updated the codebase, we've updated the AMI to include the new stuff.

Last night then, I booted up an instance from the AMI. We ssh'd in (via AWS SSM Session Manager) and checked and everything was working. Randomly about 40 minutes later it stopped pinging into our pingdom dashboard.

In trying to investigate: we couldn't start an AWS SSM session. Instead we had to ssh in via the old jump host.
On the instance intself -- the ssm-agent is not even present, not running. The custom config files (crontab and scoutdaemon) are no longer even on the system! The instance just looks like some plain RHEL from a couple of months ago, not the one we booted last night.

Has anyone seen this happen before? What's the underlying cause / what am I doing wrong?

This reduces our confidence in the stability of this