Connes's 1976 rigidity question predicted that property-TD/icc groups acting constructibly are uniquely determined by their von Neumann algebras; its embedding form (CEP) asserts every II₁ factor embeds in an ultrapower of the hyperfinite factor — a statement equivalent to many reformulations across quantum information (Tsirelson's problem) and model theory.
A foundational fixture of von Neumann algebra theory for half a century. The embedding form was refuted in 2020 by Ji, Natarajan, Vidick, Wright and Yuen via quantum complexity (MIP* = RE) — a landmark of indirect proof. The stronger rigidity expectation for specific group constructions lingered.
Disproof announced by OpenAI's Astra model, 1 August 2026, in the ten-results batch, with Lean-checked formalisation and external mathematicians preparing the manuscripts. Whatever the final peer verdict settles to, the episode already demonstrates AI operating at the frontier of functional analysis — a field long considered out of reach for language models.