The risk here is slightly overblown or misrepresented. Just because a fork exists doesn’t mean that anyone has even read it, let alone run it on their system. For this to be a real threat they would have to publish packages with identical or similar names (ie typo-squatting) to public package repositories which this article didn’t have any information on but which is a known problem long before AI. The level of obfuscation and number of repos affected is impressive but ultimately unlikely to have widespread impact to anyone besides GitHub.
Again, this existed before AI. Typo squatting, supply chain attacks, automated package uploads, CI pipeline infection, they’re all known attack vectors. That’s not to say this isn’t a concern, just that it’s a known risk and the addition of “AI” doesn’t, to my eyes, increase that risk. If your SSH keys don’t require a password, you have taken the decision to make those keys less secure but more convenient to use. That’s pretty much always the tradeoff in security.