AI model proves complex code correctness 2.6× faster than previous systems

Researchers built an AI system that can automatically verify whether code works correctly by breaking verification tasks into simpler pieces, achieving higher success rates with smaller models than existing approaches. This matters because proving code is correct normally requires expert mathematicians spending weeks on machine-checkable logical proofs — automating even part of this could reduce the time and cost of making critical software safe.

Code verification is currently a severe bottleneck for safety-critical systems (aircraft, medical devices, financial infrastructure) — formal proofs are the gold standard but take so long to construct that most code never gets verified at all. A 2.6× speedup in automated verification, especially one that works with smaller models, moves formal correctness from 'too expensive for most projects' toward 'occasionally feasible on important systems'.