"Current encryption will become obsolete" - only asymmetric crypto. Symmetric crypto, as well as hashes aren't affected (at least not much, you might want to double key sizes but you don't need fundamental changes.)
"Post-quantum key sizes are larger" - that depends. Basically, the algorithms are less efficient than conventional asymmetric algorithms, but there are quite a few different options out there with different inefficiencies. For example, SLH-DSA has tiny keys, but the signatures are huge and making them is very slow.
This is only news for the kind of people who refer to large language models as "AI".
Unfortunately, that's quite a lot of people.
.
If someone is researching AGI, there's a good chance that "whatever the hell they want" is for everyone to have access to AGI, that's the most obvious motivation for doing so. You are confusing intelligence with psychopathy. I suppose it comes from living in a society in which the media-political establishment worships obscenely rich psychopaths.
In any case, I don't know why you're so convinced that AGI would be smarter than human intelligence. I would expect the first one to be pretty basic, that's how technology normally goes.
There are two cryptographic properties here. Collision resistance means it's not possible to create two different files with the same hash. The collision resistance of MD5 is broken, which is why that attack worked. They were able to get Microsoft to sign something which contained their data, and produce some different data at the same time with the same hash. It was an attack against collision resistance because they created one file and had involvement with the creation of the other.
"Here's a file and a hash, make your own file different to mine with this hash" is a different attack against a different cryptographic property, which as far as I know is still not possible for MD5. If you just make a file by yourself, and then tell someone else its MD5, nobody will be able to make a different file with that hash and so convince this person they have the right file. The attacker needs to have some input into your file in order to attack MD5. If they do, MD5 is broken.
We are not a clone.