I personally found this question fascinating, and am surprised it didn't get more upvotes.
Here is a solution. Let $K$ be a number field, and define $K_n$ as in the question.
We can characterize membership in $K_n$ in terms of group theory. Suppose we have an element $a\in K_n$. This means that there is a sequence of algebraic field extensions $K=L_0\subseteq L_1\subseteq \ldots \subseteq L_n$, where $a\in L_n$, and each extension $L_{i+1}/L_i$ is formed by adjoining square-roots of elements of $L_i$ (and all their Galois conjugates over $K$, to keep each of these new number fields Galois over $K$). So (the Galois closure of) $K(a)$ belongs to a Galois extension $L_n/K$, whose Galois group $G$ has a normal series $G=G_0\trianglerighteq G_1\trianglerighteq \ldots \trianglerighteq G_n=\{1\}$, where each quotient $G_i/G_{i+1}$ is isomorphic to an elementary abelian $2$-group. (The converse is also true.)
Let $H\trianglelefteq G$ be the fixing field for the Galois closure of $K(a)$. We then have the normal series
$$
G=G_0H\trianglerighteq G_1H\trianglerighteq \ldots \trianglerighteq G_nH=H.
$$
This gives rise to the normal series
$$
G/H =G_0H/H \trianglerighteq G_1H/H\trianglerighteq\ldots \trianglerighteq G_nH/H = H/H.
$$
Each quotient $(G_iH/H)/(G_{i+1}H/H)$ is isomorphic to a homomorphic image of $G_i/G_{i+1}$, and hence is an elementary abelian $2$-group.
In other words, if $a\in K_n$, then the Galois group of the Galois closure of $K(a)$ over $K$ must have a normal series of length $n$ whose invariant factors are elementary abelian $2$-groups.
We know that since $K$ is a number field, there is some cyclic Galois extension $M/K$ of degree $2^{n+1}$. (Use Shafarevich's theorem on the inverse Galois group problem for solvable groups over global fields, although I'm sure a simpler argument would suffice here.) By the primitive element theorem, write $M=K(b)$ for some $b\in M$.
Since $M/K$ is Galois, with Galois group cyclic of order $2^{n+1}$, we know that $M$ can be obtained via a sequence of $n+1$ quadratic extensions. Thus, $b\in K_{n+1}$. On the other hand, such a cyclic group does not have a normal series of length $n$ whose invariant factors are elementary abelian $2$-groups. So $b\notin K_n$.