Can AI avoid defensiveness about politics?