security: which workflow works best for AI Projects?

[logic_nina]

I have a question about AI Projects that may be useful for other people working with it too.

I am particularly interested in security and accessibility. I would like to compare approaches that people have actually used rather than just theoretical suggestions.

What would you tell someone trying this for the first time?