Is what you suggest about training even possible? Most exploitation techniques are really just about having in-depth knowledge of how components work. For example, I imagine a sufficiently powerful model could fairly easily re-invent the ROP chain from first principles if it just knew how the stack works. This same principle applies to much more complex attack too; exploitation is often just an exercise in knowing vastly too much trivia, which LLMs tend to have in spades.

Remove the relevant data, and just enough of the data around it will remain that the AI will be able to close the gap if given relevant documentation.

Not to mention that those capabilities are inherently dual use. If you know how to write C safely, you know how to spot unsafe C.

Or they wanted the model to be good at these things, for the companies that legitimately need access to these capabilities.