OOP question #1 (encapsulation, coupling)

Scenario: I have an object A that has some mutator methods, and I have another object B that uses A, and that requires exclusive access to it.

Only B can use said mutator methods. If I change A externally I will break B's internal state.

$a = new A;
$b = new B($a);
$a->change(); // $b is fucked up.

What I and many others would do to solve this is:

public function B::__construct(){ $this->a = new A(); }

A is hardcoded in B's constructor and inaccessible from the outside.

Now, B could work with any object that implements A's interface, but I'm denying myself this possibility, as A is tightly coupled to B.

Question is: why we do that?

I can enforce the coupling externally if I wanted, in a factory for example:

function getSafeB(){
    return new B(new A); // A will be inaccessible externally.
}

What are your thoughts about this? Let me know in the comments.