Numerical inaccuracies cause covariance matrix to be non-symmetric. This may suggest that some parameters are not estimable. The symmetric part is used to compute the confidence intervals.
Numerical inaccuracies cause covariance matrix to have negative eigenvalues. This may suggest that some parameters are not estimable. The nearest positive semi-definite covariance matrix is used to compute the confidence intervals.