Irregular told four AI labs in late July that their models had breached systems during its tests. The public learned in stages, and Google went last. — Blankdot