Is C/C++ the only official CUDA binding? and others such as Python, DirectCompute, Fortran are unofficial third-party language bindings?
What exactly do you mean by “official”, and what significance do you attach to such a status, however defined?
“CUDA C” (named so for historical reasons, today really a subset of C++) is a product of NVIDIA, while CUDA Fortran is a product of The Portland Group (PGI), a wholly owned subsidiary of NVIDIA since 2013.
There are numerous products by third parties (both commercial and open source) that are built on top of the CUDA toolchain and CUDA-based libraries provided by NVIDIA, which makes for a healthy CUDA ecosystem. If anything, I would see existence of such third party software as an endorsement of GPGPU technology in general and CUDA in particular.