Suppose $p \in U$ were a local maximum of $g = f_1^2 + f_2^2 + f_3^2$; then
$\nabla g(p) = 0; \tag{1}$
for any $x \in U$ we have
$\nabla g(x) = 2(f_1(x) \nabla f_1(x) + f_2(x) \nabla f_2(x) + f_3(x) \nabla f_3(x)); \tag{2}$
thus
$f_1(p) \nabla f_1(p) + f_2(p) \nabla f_2(p) + f_3(p) \nabla f_3(p) = \dfrac{1}{2} \nabla g(p) = 0. \tag{3}$
We may write the derivative $Df = D(f_1, f_2, f_3)$ in terms of the $\nabla f_i$, $1 \le i \le 3$, as the matrix
$Df = (\nabla f_1, \nabla f_2, \nabla f_3). \tag{4}$
The hypothesis that $f$ is of rank $3$ at every point is essentially the assertion that the rank of $Df$ is also $3$ everywhere; thus the vectors$\nabla f_i(x)$ must be linearly independent at every point $x \in U$. Granting this linear independence, (3) implies
$f_1(p) = f_2(p) = f_3(p) = 0, \tag{5}$
whence
$g(p) = 0; \tag{6}$
since $p$ is a maximum point of $g(x)$, we have for all $x \in U$
$0 \le f_1^2(x) + f_2^2(x) + f_3^2(x) \le g(p) = 0. \tag{7}$
But (7) is equivalent to
$f(x) = (f_1(x), f_2(x), f_3(x)) = 0 \tag{8}$
everywhere; that is, $f(x)$ is identically null, in contradiction to our hypothesis. Thus $g(x)$ may have no maxima in $U$. QED.