The most infamous case is that std::unordered_map is essentially required to be a bucket-based hashtable, whereas it's usually better to have a probe-based hashtable instead.
They feel like a 1980s feature, if there was a hash table in C's stdlib, this is the hash table you'd expect. But it's not in C and it wasn't from the 1980s, it's from the same year Osama Bin Laden was killed by US special forces.
Rust existed at the same time this data structure was standardised. Not Rust 1.0, but that's insane, it's like when you realise millions of older Spanish people were alive before their country was a democracy. Many are old enough to remember the prior regime.
The C++ 11 standard was not required to standardize broken garbage from many years ago.
WG21 pulled C++ 0x Concepts (a much more powerful feature than the "Concepts Lite" which landed in C++ 20) because they argued it might take too long to implement, ridiculous as that claim is, so there's no reason to keep something terrible just because it's in a draft paper.
It's hard to even believe this was trying to be general.
Notice that not only the hash tables themselves, but the infrastructure is garbage too. In 1998 C++ STL is one of the earliest generic programming APIs, nobody knows how you should implement the generality of something like hashing, fine.
But by 2011 people had tried several things, and this (the std::hash templated callable which returns a machine-word integer) is not even a reasonable attempt at a known good design.