paper / thinkingmachines / May 15 / failedSafeWork-R1: Coevolving Safety and Intelligence under the AI-45$^{\circ}$ Law