 Solution

The claim is not implied. The intervention setting $A=a$ removes the outgoing dependence of descendants on the observed value of $A$, but the bidirected path $A\leftrightarrow Y(a)$ remains open. Conditioning on $X$ does not block latent confounding between $A$ and $Y$.