https://betterwithout.ai/cybersecurity-vs-AI David Chapman's advice on how to make software secure. note that all of these are R&D projects that are not ready for wide deployment yet.
https://en.wikipedia.org/wiki/Stem-cell_therapy stem cells are in routine use for blood cancers, and are in the clinic for neurodegenerative diseases, spinal cord injuries, frailty (!), recovering hearing and sight in some kinds of deafness and blindness, and various orthopedic and wound healing applications
https://www.thediff.co/archive/the-economics-and-politics-of-pacing-the-frontier/ Byrne Hobart doesn't give strong opinions on the "pacing the frontier" thing, except to note that the labs' proposal might be good for them economically, so long as the regulator is relatively independent (like the Fed) and not responsive to public opinion
he says eudaimonistic virtue ethics is about "doing X X-ingly", i.e. promoting art artistically, pursuing honesty honestly, etc.
to consider X a "value" in a virtue-ethical/eudaimonistic framework requires believing that you promote X by practicing X. you would not trade X today against X tomorrow. you would not do bad things for good ends.
the nasty things we worry AI might do usually involve doing something we would not approve of, instrumentally as a means towards some end we did assign. or extrapolating a goal we assign in a way we no longer approve of. This is mitigated if the AI is "virtuous" or always tries to "do X X-ingly."
https://arxiv.org/pdf/2608.03958 game theory of "embedded Bayesian agents" that cooperate with each other when they infer others are similar.
https://www.agi-institutions.org/ wiki of ideas of post-AGI institutions that need to exist. (in a multipolar world with both humans and autonomous AI agents.)
https://arxiv.org/pdf/2607.02755 empirical research by Elliot Thornley that you can train AIs to be risk-averse in small-scale situations and they'll generalize to large-scale situations