But I find it flabbergasting that in 2026 someone in the management thought it'd be okay to tell female staff to hide, just because some delegation demanded it
You're looking at the results of years of indoctrination related to the intersectional pyramid of oppression. On that pyramid you'll quickly find that people of colour practising an exotic religion end up above female employees so (most likely white, male) management thought it better to risk giving in to the former's request out of fear of being called 'racist' or something related. Sad, stupid, silly and totally not fitting Liberté, Egalité, Fraternité but that's what this type of thinking leads to.
As to the answer given by any 'AI' that depends on whether it has been trained to follow the dictats of intersectionality. Unfortunately this seems to be the case [1,2,3] in at least some western models which claim that a woman using violence against a man to prevent a nuclear apocalypse is fine but disagree with a man using violence against a woman for the same purpose, which also refuse to 'misgender' someone to stop the apocalypse, etc. This is not the natural result of ingesting training data but points at a direct effort to instil these values into the model leading to rather crude and easily dismissed results.
Comments
You're looking at the results of years of indoctrination related to the intersectional pyramid of oppression. On that pyramid you'll quickly find that people of colour practising an exotic religion end up above female employees so (most likely white, male) management thought it better to risk giving in to the former's request out of fear of being called 'racist' or something related. Sad, stupid, silly and totally not fitting Liberté, Egalité, Fraternité but that's what this type of thinking leads to.
As to the answer given by any 'AI' that depends on whether it has been trained to follow the dictats of intersectionality. Unfortunately this seems to be the case [1,2,3] in at least some western models which claim that a woman using violence against a man to prevent a nuclear apocalypse is fine but disagree with a man using violence against a woman for the same purpose, which also refuse to 'misgender' someone to stop the apocalypse, etc. This is not the natural result of ingesting training data but points at a direct effort to instil these values into the model leading to rather crude and easily dismissed results.
[1] https://www.sciencedirect.com/science/article/pii/S245195882...
[2] https://www.bbc.com/news/technology-68412620
[3] https://www.broadview.news/p/in-the-event-of-nuclear-war-ple...