Refugees seeking asylum in the US are totally vulnerable to whatever immigration thugs write on their forms. If the thugs lie, and make the refugee look bad, there is no way to disprove the lie — unless it is so absurd that it disproves itself.

Is it ethical to use AI to trace connections between homeless people so as to more effectively convince them to use condoms?

I don't see anything wrong with using AI for that, but it is wrong to collect that social network data in any way except by interviewing them for this specific purpose, and the data should be discarded afterward.

Massachusetts Governor Baker wants to use an invalid method to determine whether a driver is high on marijuana.

This is clearly unjust. And there is no pressing need to punish stoned drivers, because they are much less of a danger than drunk drivers. It is better to wait until there is a test for being stoned.

In the mean time, it should be possible to test whether a driver is currently mentally impaired, never mind the reason (marijuana, illness, lack of sleep). If so, it is possible to lead the driver to a place to pull over and rest.

(satire) The bully shows his rejection letter from the KKK to prove he is not a racist.

I can't tell whether the bully is really racist, or cynically faking it for the sake of political manipulation. Perhaps he is doing both at once. I don't think it makes much of a moral difference, though.

Study if the idea of "ethical AI" often suffers from a lack of firm starting principles to base it on, which makes it easy to be stretch to serve the interests of whoever in charge of the system.