For example Padding Oracles are not an inherent flaw in TLS 1.2, the protocol itself doesn't oblige you to tell attackers "You guessed wrong" until they guess right. But it becomes your responsibility to not let them know, an apparently optimal TLS implementation functions as a timing Oracle for CBC mode and can be exploited. Making CBC implementations sufficiently opaque that an adversary can't tell if they guessed the padding bytes correctly is really hard, OpenSSL have failed twice so far already. In TLS 1.3 this isn't an issue, it only has AEAD primitives.
And lack of Forward Secrecy is a pretty serious problem for some applications. It's not fatal, but it's a problem. In TLS 1.3 you always get Forward Secrecy. In TLS 1.2 it was optional, so unless you're explicitly configuring the ciphersuites used there's no guarantee.
Yet. That we know of.
> yay, instant downvotes! I hope you enjoy being herded by Google
Yay! Complaining about downvotes! I hope you enjoy more downvotes.