Research Do Corporate Incentives Teach AI to Protect Itself? July 13, 2026 · 4 minute read The tools that make an AI safe, like kill switches, also make it stoppable. A model therefore has an incentive to help limit rival AIs but not itself.
"Surrounded by Enemies" (Kringsatt av Fiender), or "To The Youth" (Til Ungdommen) English Translation (48 recent reads) Audiovisual reading (27 recent reads) Mediocre Intelligence, Outstanding Results (24 recent reads) AGI Security: Rethinking Society's Infrastructure (11 recent reads) Cybermorphism (10 recent reads) To An 18 Year Old Me (9 recent reads) Go on a yearly hiatus (8 recent reads) Vibe Coding: Taste Is The Last Frontier (7 recent reads) An Athlete's Mind (6 recent reads) Don't Give Up (6 recent reads) Let's make AI agents safe (6 recent reads) The Designer, The Craftsman, & The Programmer (4 recent reads) "Is The Light Merely For The Learned?" (Er Lyset For De Lærde Blot?) English Translation (4 recent reads) Sentware (4 recent reads) How to host a hackathon (2 recent reads) AGI Privacy (2 recent reads) Raising for the Endgame: An AI Safety Founder’s Primer (1 recent reads) The AI Bubble (1 recent reads) The AGI endgame (1 recent reads) My Tools (1 recent reads) The Second And Third Falling (0 recent reads) Learning From Architecture (0 recent reads) The concrete risks of AI misdeployment (0 recent reads) Confident optimism (0 recent reads) The Expensive Internet hypothesis (0 recent reads) "Just do the thing" (0 recent reads)

Research

Socials

News