Let me add some details to Nicholas Kuhn's answer (mostly for my own edification).
First, since Steenrod squares are stable cohomology operations, it is enough to show that the maps $\operatorname{Sq}^I : H^{2^j-2k}(\mathbb{CP}^n) \to H^{2^r-2k}(\mathbb{CP}^n)$ must all be zero.
Lemma: The subspace $F_p = \operatorname{span}\{y^m \mid \alpha(m) \leq p\}$ is a sub-$A$-module of $H^*(\mathbb{CP}^n)$; here $A$ denotes the Steenrod algebra.
Proof: We just need to check that $F_p$ is preserved by each Steenrod square. For degree reasons, we have $\operatorname{Sq}^{2\ell+1}(y^m) = 0 \in F_p$, while $\operatorname{Sq}^{2\ell}(y^m) = \binom{m}{\ell}y^{m+\ell}$; see this answer. If $\binom{m}{\ell} \equiv 0 \bmod 2$, then $\operatorname{Sq}^{2\ell}(y^m) = 0 \in F_p$. On the other hand, if $\binom{m}{\ell} \equiv 1 \bmod 2$, then $\alpha(m + \ell) \leq \alpha(m) \leq p$ as pointed out by Tanaka, so $\operatorname{Sq}^{2\ell}(y^m) = y^{m+\ell} \in F_p$. $\quad\square$
Now let $p = \alpha(2^{j-1}-k)$.
Note that $y^{2^{j-1}-k} \in H^{2^j-2k}(\mathbb{CP}^n)$ is an element of $F_p$, while $y^{2^{r-1}-k} \in H^{2^r-2k}(\mathbb{CP}^n)$ is not because $\alpha(2^{r-1}-k) > \alpha(2^{j-1}-k) = p$. By the lemma, we see that $\operatorname{Sq}^I(y^{2^{j-1}-k}) \in F_p$, so $\operatorname{Sq}^I(y^{2^{j-1}-k}) \neq y^{2^{r-1}-k}$ and hence $\operatorname{Sq}^I(y^{2^{j-1}-k}) = 0$, i.e. $\operatorname{Sq}^I : H^{2^j-2k}(\mathbb{CP}^n) \to H^{2^r-2k}(\mathbb{CP}^n)$ is the zero map.