The ability to call the name method suggests that, similar to has_many :accounts, class_name: Account.name, an offense should be registered:
https://github.com/rubocop/rubocop-rails/pull/1180/files#diff-9d8842c039d0144f6548e139d525048bb9e0651ba63906ca875e35ff52140498R16-R25

I understand the purpose of this cop to be to avoid the use of constant references, which lead to loading all the referenced classes.

Account.name will resolve Account, some_class_in_a_variable.name will not load the class (it will already have been loaded).

I can see how this is kind of an edge case though since it does seem to require the loading of a class (unless the name method here is not Class#name).

I wonder - isn't your concern covered by it 'registers an offense when parameter value is a local variable assigned a constant'?

I can look into this, change this test case to use different variable / method names and add another test case for

expect_offense(<<~RUBY)
  the_class = Account

  has_many :accounts, class_name: the_class.name
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^ Use a string value for `class_name`.
RUBY

(Although I think this makes me think that the message shouldn't be "Use a string value for class_name", but rather "Don't use a constant reference for class_name")

OK, I tried it -

  the_class = Account

  has_many :accounts, class_name: the_class.name

does not register an offense even on main. So for now I've just changed the variable and method name in this test. I believe the behavior of the cop is as expected except for the above example which didn't work before either. But it's better to have a false negative than a false positive.