These safety tests are, humorously, that it isn't censored enough. That it doesn't stop you from talking about things that the arbiters have decided you shouldn't be able to talk about.

Which makes it rather ironic given the normal "...but tiananmen" retort.

Put Gaza or Israel or Palestine into your Western AI models...

“Safety test” sounds like things a self-appointed group of experts has decided people shouldn’t talk about.

Thank you Deepseek for pushing the sector forward and exposing the scammers. The world owes you.