Authorize.net uses xml.dom.minidom to parse their responses. minidom doesn't handle unicode. We must pass it an encoded (as in bytes, not codepoints) string to it. Luckily, authorize.net has a hook to run code after execution.

fix-authorizenet-unicode-errors.py

from authorizenet import apicontrollersbase

def afterexecute(self):
    self._httpResponse = self._httpResponse.encode('utf8')

apicontrollersbase.APIOperationBase.afterexecute = afterexecute