Every July, the brightest teenage mathematicians on Earth sit down to the International Mathematical Olympiad, six problems over two days, and most of them do not finish. The IMO is not an arithmetic test; it is an exam in invention, where each problem demands an argument nobody has seen before and a single question can defeat all but a handful of the world's most gifted young minds. So when Google DeepMind reported that an artificial intelligence had solved these problems at the level of a gold medallist, it landed as something more unsettling than another benchmark. The machine had done the thing we thought was ours: it had reasoned its way to a proof. The interesting part is not that it won. It is how, because the how is the difference between a calculator and a mathematician, and this was not a calculator.