The usual response is to ask what's the share of modern FORTRAN compared to old FORTRAN. If most of the people working with FORTRAN work with the old one and don't modernize it, then the reputation is deserved.
People don’t modernize their codes because the Fortran compilers allow them to, and the Standard is reluctant to delete old features (and compilers would not delete them anyway).
Unfortunately, that means that old bugs can lie undetected. I don’t think any other language/compilers have squared that circle.
And there’s plenty of new projects started in Fortran.
-But we can build with concrete and steel and plastics and all sorts now.
-Most people use bricks though. And people are not tearing their brick houses down to rebuild them in steel. See? Building is old fashioned, antiquated!
In fact, nobody now starts a new project in F77. There’s a lot of F77 brick code around, yes. And it still does the job it was asked to do even though it gets moved to a new district every ten years.
! MARK 1 RELEASE. NAG COPYRIGHT 1971.
It has been revised since then to use free form, structured programming, MODULEs, KINDs, INTENTs and IMPLICIT NONE.
If we can do it, everybody can. I have already mentioned some reasons that people don't do it. None of the reasons I have heard are "Fortran is too old and clunky".
As for which version is used in the wild, it's the same as with many other languages that have stuck around for a long time. Plenty of legacy code which might still be F77, but practically all new projects, or extensive modernization of legacy code, will make use of the modern features.
I suspect Fortran's reputation is largely due to many of the foundational libraries of numerical computing like BLAS and LAPACK, are F77 (LAPACK uses a small number of 'modern' features, but largely it's still a F77 codebase). This is the kind of Fortran that non-Fortran programmers might come into contact with. But the Fortran that Fortran programmers write today is definitely much more modern.