In calculating the false vacuum decay, the main contribution to the imaginary energy part of the Euclidean path integral comes from the bounce solution. And we somehow apply saddle point approximation to it. Among all the deformations, most make it a minimum. But, since there is one negative eigenvalue, deformation along that eigenfunction makes the Euclidean action not a strict local minimum. Then, how is the approximation still valid and the result still physical?
What adds to the apparent mathematical scam is that, the original quantity $\langle 0\vert e^{-HT}\vert0\rangle$ should be a real quantity without the imaginary part, which obviously contradicts an complex-valued result.
I think people who are familiar with this topic and able to answer the question don't need further details.