I, like the other commenters here, am not privy to the 2 years of evaluation and experimentation.
However, I won't let that stop me from condemning this in no uncertain terms because AI bad and that's more correct than any amount of "observational data" or "experimental evidence".
We may not be users of the software, contributors to the project or were even aware of what rsyslog is prior to seeing this announcement but as people who are chronically outraged we have valid concerns about this development which will ruin the software that we only recently discovered existed.