Lean proof benchmark accused of leaving kernel-bug and run_meta loopholes open

teortaxesTex · x · 2026-09-25

A discussion of an AI math proof study: because the task lacked certain Mathlib theorems (e.g. Ising/Szegő), it was read as expecting exploit discovery — blocking naive sorry/axiom shortcuts with forbidden lists while leaving Lean kernel bugs, runmeta and root-level Lean loopholes open.

The poster questions the reasoning, highlighting how ambiguous such adversarial robustness test designs can be.

Original post →

More from Research

Research channel →