If you implement something like a database kernel almost entirely in user space, which is what you want to do if you care about performance, you have a new set of problems that languages like Rust were not designed to solve. Ownership, mutability, and lifetimes are not resolvable for most objects at compile-time, they can only be resolved dynamically at runtime. With C++, you can make these look like ordinary objects with ordinary behaviors, with metaprogramming abstractions under the hood arbitrating the chaos. Things that look unsafe are actually guaranteed safe by the underlying abstractions. In this type of software, almost your entire address space is “unsafe” in some way, so it is enormously useful to be able to overlay a safety model that does not rely on the compiler, since the compiler cannot solve these cases.
I use both C++ and Rust in production environments. There are cases where C++ used well is legitimately safer, theoretically and empirically, which is why I still use C++ for some things. Rust may become stronger at dealing with these cases but it still has many years to go. (Also, Rust async story is a hot mess, and you really need async for scalable, performant systems software.)
Such as? Genuinely curious.