It works moderately well in securing against a program going rogue and trying to prevent the user from moving the data on their device to another program serving them better. (Or underrated but important, making it less painful to enter and re-enter information in say a dark pattern subscription cancellation form.) Unfortunately it's easily defeated as a security mechanism, but it's still way better than nothing.

What would you do instead?

One simple solution is just to gate the programmatic access by confirming with the user when the app wants to read the clipboard.

Your comment is bordlerine absurd given the utility of such a thing as copy and paste. If it is satire, then you got me.

Today's paranoid times?

The current status quo for software architecture is giving a swarm of LLM agents root and full access to user data and letting them run amok, unsupervised, connected to a proprietary remote black box service either being monitored by Palantir or the CCP. What is this "paranoia" you speak of?

Could make a copypasta out of that (ha, ha.).

But like that genuinely applies to a quite large amount of things. Screenshots, screen recordings, actual regular physical cameras, unencrypted physical cables, publically-usable GPS, microphones, etc. (these are even reasonably-realistic - Wayland & phones have restrictions for screen(shots| recording), backslash on cameraful glasses, DRM, criticism of airtags, leaking typed text through the typing sounds)

Could even go broader and count in people being non-anonymous online, and people being anonymous online, as both being bad for security; were the world to have settled on one, the other would be utterly crazy.

Giving semi-arbitrary anything to LLMs should absolutely very-trivially count in this too, except that for whatever reason it's very often just not? Clipboards would get a pass just fine.