DARPA has launched its “Exponentiating Mathematics” (expMath) initiative, a bold move to accelerate mathematical discovery using AI. The goal: build AI “co-authors” — systems capable of breaking down complex proofs into manageable lemmas and composing novel theorems. DARPA is seeking proposals combining large language models, reinforcement learning, and program synthesis to advance symbolic reasoning in mathematics. The roadmap is modular: one track targets AI development itself, while another evaluates AI against professional-level math challenges. If successful, expMath could radically shorten research timelines, impacting cryptography, materials science, fluid dynamics, and more. The program also seeks to democratize access to advanced math tools. DARPA’s ambition is clear: redefine how humans and machines collaborate on deep mathematical inquiry.