I've been thinking about AI doom scenarios for many years. I've always been a skeptic — paperclip nonsense and all that. But thanks to the Coldcard hack, and a recent reading of Ronald Wright's A Short History of Progress, I kind of see it now. (Great read by the way; crisp, well-argued, fascinating, crafted with care). Civilization is precious, but fragile. It can die when facing unknown unknowns or exponential trends that catch us unawares, until it is too late. Arms races tend to be bad for civilizations too. One known unknown when it comes to AI is this: whether it will ultimately create any attacker/defender asymmetries, or individual/institutional asymmetries. One factor here is just how far the open weight models can go, and whether they'll be permitted by authorities. Alas, I can't write out the unknown unknowns. But I'm now more sure than ever that they're out there. I'm an optimist at heart — unlike Wright. But I now see connections between my more conservative impulses (watch for Chestertonian fences; beware rapid change) and what is actually happening with AI and cybersecurity. This is only the beginning. What a time to be alive and to be monitoring the situation!