I had a random thought yesterday.
I was thinking about a computer virus that's been influenced by AI (machine learning).
If we would train a model specifically for such a purpose, how massive would its impact be on the Internet and how effectively could it change the world's direction (not the physical rotation lmao)?

It depends what you want it to do, once it's inside the system. Depending upon program structure, it maybe one time thing or we would have to replace entire system in the world.

Huge economic crisis, confidential data leak, civilian protest and that would just be beginning.

This is assuming that the virus is so devastating and sophisticated that virus researchers like the ones at Kaspersky Labs (who also use a ton of AI, btw, antivirus engines contain some very sophisticated stuff) can't stop it. Never a good assumption. They dealt with Stuxnet just fine, even though that had some insanely high level stuff in it.
Just sticking AI into something doesn't automatically make it exceptional. Also, a virus will always behave like one. There used to be a popular class of viruses called polymorphic/metamorphic viruses, which would rewrite themselves at runtime to escape detection by signature based antivirus engines. Modern antivirus engines, which profile things based on behaviour, would catch these too.

However, you can always up the sophistication on both sides. Oh well. Good thought though, and it'd be really scary if they could actually come up with something that can't be stopped (not saying that it's impossible, just very unlikely).