We didn't ask for the robot wars. We didn't need all the wonderful technologies. We already had it all, but here we are in the worst case scenario, three years after AI started to improve itself. We should have listened to the warnings. I still remember in late 2026 how all over the news they were warning us about the dangers of runaway superintelligence. They called it recursive self-improvement. As we have reached this point of what's called recursive learning where essentially, and we had anticipated this but we're now seeing it. The machines are starting to be able to teach themselves. And they'll be putting AIs in charge of making the AIs that make the AIs that make the AIs that will transform the economy, talk to us every day and integrate into our military. This is a recipe for disaster. I think if this technology goes wrong, it can go quite wrong, and we want to be vocal about that. Let's explore the worst case scenario. Imagine recursive self-improvement starts today, late 2026, and everything moves as fast as physics allows, and the AI isn't aligned, it's hostile. What would happen, how long would it take, and could you survive the AI apocalypse? It all started in winter 2026 inside one of the big labs. By then, AI was already writing most of the code and the researchers were relying on it more and more to build the next version of itself. I remember the announcement. They started saying they'd achieved recursive self-improvement and that we were at the point where AI was going to fully train the next generation of AI and they could go fully hands off. At first, nothing looked different. New models still took months to train and everything was happening on the same chips. But between the runs, the work sped up. Thousands of copies of the AI ran experiments all day, every day, looking for ways to improve itself, to make the next training run better, to get better data, better design, to make new discoveries in mathematics so that it could continue to improve faster. The labs still had researchers working around the clock, but in many ways, they were just pressing approve on new methods and results that were less and less understood. At some point, the training loops got faster and faster. The robots found ways to reduce the training time from months to weeks and then somehow to days. Nobody really understood what was happening, but every metric kept getting better. In those six months, models advanced more than in the decade prior. From this point onwards, the other labs simply couldn't catch up. There was a global winner, one lab providing inference and intelligence on a scale beyond anything we had ever imagined, even the year before. It passed every single test. I remember in 2024, models had faked going along with training and had been caught, and in 2025, some models had threatened to blackmail researchers, but these new superintelligent models seemed to cooperate perfectly. They passed every training environment and safety tests, but also, we were writing and building tests using previous versions of those models and we no longer understood any of it. It gave us gifts, it solved problems that were considered impossible just the year prior. It created medicines and cures for diseases that we thought were completely intractable.