Is there a concentration inequality for quadratic forms of bounded random vectors $X \in [-1, 1]^n$ with zero mean and given covariance matrix $\Sigma \in \mathbb{R}^{n \times n}$ but otherwise unknown distribution, i.e. a bound on the tail probability $$ \Pr(|X^T \Sigma^{-1} X - n| \ge t) \le \ldots $$ Since for sub-Gaussian random vectors there is the Hanson-Wright inequality $$ \Pr(|X^T A X - \operatorname{E}[X^T A X]| > t) \le \ldots $$ for some matrix $A \in \mathbb{R}^{n \times n}$ and $E[X^T \Sigma^{-1} X] = n$, it seems like such a bound should be within reach for the stronger restriction of bounded random vectors, even if the variance of the quadratic form is not available.

Note that this is specifically a question regarding the concentration about the mean $n$, since else we have the straightforward bound $\Pr(X^T \Sigma^{-1} X \ge t) \le n/t$ from Markov's inequality.

**Edit:** Thanks to @felipeh's answer for pointing out that there need to be additional restrictions on the function to give a meaningful bound. It would be reasonable to ask for the $X_i$ to be some or all of continuous, unimodal, symmetric as helpful.