To be completely fair, even open-source AIs are a little bit of a black box due to the way neural networks work - but I'd greatly appreciate if we at least knew the parameters on which it is trained.

It is absolutely possible to train all sorts of biases in a closed-source AI, and that's what would be very hard in an open-source model. You can roughly set up outputs at whatever. In other ways, using open-source practically removes the malicious human factor (without removing positive impact)

Open-source models also can't be restricted, paywalled or limited in any meaningful way, which is also vital.