Using text-to-speech software and a cron job
At random intervals (several times an hour):
Get volume level
Lower volume level (think ghostly whisper)
Use T-T-S to say something barely audible
Return Volume to previous level

Bonus points if your hands are away from the keybd, or it’s someone else’s machine

someone at work got me once by changing my .bashrc to include a file that would sleep 0.001 on every command, but every time the file was sourced it would edit itself to increase the time by 0.001 seconds.

A week later I was about to toss my computer off the roof.

fortune cookie -s | cowsay_random | xmessage -xrm "*message.scrollVertical: Never" -buttons "" -nearmouse -bg gray -fg black -font 7x13bold -file -

Requires that fortune be installed along with the eponymous cookie fortunes set (these usually come with the default install).

cowsay_random is not installed fully working by default but can usually be found among the files installed with cowsay. (The #! header might need to be updated to use python2 instead of just python as I found out when testing this)

xmessage is a seriously old program but it works just fine. The command line supplied pops up the output of the preceding commands near where the mouse is on the screen. The -file - bit is necessary to read the pipeline.

See the man page for xmessage if any of the other options aren’t too clear.

Your further mission is to arrange for this to trigger somehow or another, be that at random or when some other event occurs.

Caveats: 1) No idea if this works on Wayland. 2) No ANSI colour code support, so no lolcat. Figure out a way to use an xtermor something if that’s what you want.

In university we used to prank each other by accessing less knowledgeable colleagues’ machines via ssh and running random commands. We would drive them nuts by ejecting the CD tray, or playing random wav files, or popping up silly websites. The lan was not really secure, neither were the passwords.