Flagging Malware is hard, and research/dev tools are always behaving at least similar to Malware (because we want to get data/do stuff regular users won't do).
making an exception for such a heuristic is, in all cases, wrong since it will always be abused.
The actual answer is: Defender needs a PUP category.
Nirsoft tools? Bam, "virus" and "malware". How dare you!
Tailscale website? Uh-oh, ZScaler thinks that's a "remote access tool" so you're being given a click-through formal warning!
The Framework website? Uh-oh, .work is a bad TLD! Can't browse to that, it could be evil!
Like Powershell, or Microsoft Automate or Tosca, who can all run keystroke injection, but aren't flagged.
Then, when the exception has to be revoked, the backlash is massive. Look up the recent example of the driver FanControl used to issue SMBus commands being blacklisted.
It has been the norm since we first started automating processes designed more for people than automation. It will remain the norm for as long as that exists.