This is a question I ponder about a lot. How much freedom should you give your students in their work and how much should you trust them that they stick to the rules of scientific working even when there are so many tempting options that could make their life easier.
GMP wrote this great article about taking over paper drafts that don't come along good enough or fast enough if written by students. This is one great example about the boundary between giving freedom and trusting that the students will do a great job - even though it might take a while - and taking over control at some point to get that paper out to the public.
Another related point is plagiarism. We all would love to have students who don't copy and paste sentences or whole paragraphs from other peoples work and it would save so much time to not have to check at all. I often feel bad, that I do check my students works for plagiarism, because -hey- they are not little kids anymore, they are grown-ups in their mid 20s, they know the rules, they understand why these rules are in place. But often enough I get reports written in more or less heavy accent of whatever nation the author comes from and in between there are these flawless, well constructed sentences. Pasting them in Google gives a straight hit in 99% of the cases - very sad, but at least you know that they had a look at some of the famous papers in their field. So I continue to check on them, which is annoying, takes a lot of time and does not build my trust in their general scientific work.
Because if they copy and paste from other people for their reports, what is their standard for their own work? How well do they conduct their experiments? Do they just show me the one flawless data set and don't mention the other 10, which look "unexpected"? I had a case like this once and it made me dig through the persons raw data, which was a real pain. Of course there are people with very high standards, who prefer to measure 5 additional data sets just to be sure and who are not afraid of data that does not look like textbook but instead ask the big "why is this" question. If I'd dig through their raw data I'd understand that they would feel treated like little kids.
Another sensitive topic is the communication with collaboration partners. I'd love to have my students discussing their projects with more senior scientists and this means that there has to be an initial contact at first. If they don't meet in person on a conference or during a visit, this contact will be most likely via email - and there are so many pitfalls to be aware of. Some students just write perfectly polite emails just by themselves, but others come across quite rude - even if from their cultural background they should be very good in expressing respect towards a more experienced person. Does that mean I have to check emails before they are sent? That again feels like massive paternalism.
How do you determine the level of control a certain student needs? Do you fully trust them in the beginning and adjust the control level on the go? Or do you control a lot and let them run free if they seem to get the concepts of scientific standards? What if they don't care and try to take the "easy path" again and again?