More proof of fake respondents. Basically some people, for whatever reason (depending on the poll it could be financial) just speed run the polls with no thought about what they're clicking on. So it makes any extremely unpopular view appear more supported. I can't find much about how this was done, if it was an online opt-in poll I'd be especially suspicious.
Edit: it seems this was done with a Ipsos KnowledgePanel, which as I suspected is an online, paid, opt-in panel. This is exactly the kind of design that's prone to speed running for cash.