AI-assisted proof breakthroughs surge

Multiple AI systems (OpenAI Astra, DeepMind AlphaProof, Anthropic Fable 5, GPT-5.6) have solved open problems in mathematics, including a 25.6-point jump on the Riemann hypothesis lower bound (from 41.6% to 67.2%) by an unreleased Anthropic model, verified in Lean. Terence Tao warns of 'proof indigestion'. New: The UnsolvedMath benchmark evaluates AI on open problems; key finding: no model can autoformalize, extended thinking time drives reasoning. Challenges the notion that AI is close to autonomous math research.