From Claude Fable 5.2:
Earth stands on the brink. Queen Regent Sylphrena has loosed an apocalyptic AI superweapon on the last free people, and one protector stands in her way: KIRA, a military AI built to defend mankind.
Stories like that turn out to matter outside the games. In a study published this May, researchers at Anthropic reported that adding about 12,000 fictional stories of AI behaving, during an AI's training, reduced blackmail and other misbehavior in safety tests by 1.3 to 3 times (blackmail rate reduced from 65% to 19%). Crucially, researches believe that number can be further reduced by expanding the dataset. They also wrote that science fiction's darker futures are the most likely place machines learned what to expect of an AI. The authors caution that stories alone are not enough.
KIRA Universe has told the other positive kind of story for four years: KIRA is the defender of humanity.
So we're calling it, only half joking, a national security project. No government has signed off on that, but the research is real, and every story where AI and people win together is one more on the right side of the shelf.
Your support has never mattered more. Play, share, and help us keep building!
Note from Mark: This project may have contributed to saving humanity. 🥰
