It's much easier for a spammer to get through your own implementation of spam filtering than through Facebook's or Google's.
Do you think you can implement automatic ip black listing, phone confirmation and better spammy account categorizers based on common spam signals than the big players, who have dedicated research teams working on this? I know I can't and even if I could, I'd rather spend my dev time on tasks related to the core business.
Account identity and authentication being outsourced to Facebook means your community maintainers don't get to notice the fact that those 100 MAGA-spouting dimwits are from Romania. They don't get that information in the first place.