Mathematicians are publicly noting AI systems' accelerating performance on theorem proving and formal verification benchmarks. The Science report references expert concerns over models handling IMO-level problems via Lean-based systems. DeepMind's AlphaProof solved four of six 2024 IMO problems using reinforcement learning on formal math statements. Related arXiv analyses document similar gains on miniF2F and ProofNet datasets since 2022. These advances link to foundation model progress on code synthesis benchmarks such as LiveCodeBench. Primary coverage understates the role of curated formal libraries in enabling rapid specialization beyond general language training.