The company says Claude Opus 5.5 scores well on alignment measures, but that “public policy” will eventually need to step in to protect the world from recursively self-improving AI.