I wonder how much of that sort of thing is driven by having trained their models on their own internal codebases? Because if that's the case, careful and defens

I wonder how much of that sort of thing is driven by having trained their models on their own internal codebases? Because if that's the case, careful and defensive being the default would be unsurprising.