Training LLMs that are both safe and accurate presents challenges such as carbon footprint, fine tuning, accuracy on main tasks, and not having easy work arounds to avoid safeguards.

This work reminds me greatly of the book: Taming Information Technology which was written in response to the proposal to automate system administrator tasks.

ICS are also under lots of regulation.

Increase self-efficacy. <- mirrors my own experiences that are hard to tell if it is even working.

If security and privacy was a character, what traits would they have?

Lots of great discussions around yours of errors in papers. Sadly statistics errors are common.

Security keys had a range of usability challenges linked to their physical nature, such as having to carry them and find a comparable USB slot for them.

Primary password typing meant that participants saw no improvement in time use. Why type a password just so the software could enter a password. Easier to just enter the password directly.