Let $A,B\in\mathbb{R}^{n\times n}$ be two nonsingular matrices with $A\ne B$, and consider the following partitioned matrix $$ M:=\begin{bmatrix}AA^\top + BB^\top & A^\top \Delta_1 A + B^\top \Delta_2 B \\ A^\top \Delta_1 A + B^\top \Delta_2 B & A^\top A + B^\top B \end{bmatrix} $$ where $\Delta_1$ and $\Delta_2$ are diagonal matrices with $0$-$1$ diagonal entries.
It is quite easy to see that if the diagonal entries of $\Delta_1$ and $\Delta_2$ are $1$s, then $M$ is singular. However, numerical simulations with randomly generated $\Delta_i$ (whose diagonal entries take values $0$ or $1$ with probability $1/2$) suggests that, in general, it is very "unlikely" that $M$ is singular. I'd like to understand if there is a way to formalize this fact, so my question is:
Given all possibile choices of matrices $\Delta_1$ and $\Delta_2$, how is it "likely" that matrix $M$ is singular?
I'm aware that my question is not very rigorous, but I hope it's understandable.
Thanks in advance for any help.