An event has occurred in the world of cryptography and fundamental mathematics that could change the approach to solving the most complex problems. Dmitry Rybin, a graduate of the Higher School of Economics specializing in network flow theory, has managed to disprove the Dinitz—Garg—Goemans conjecture, formulated back in 1999. For nearly 30 years, this conjecture remained unproven, considered one of the key hypotheses in its field.

Remarkably, the tool for this breakthrough was not human intuition, but the language model GPT-5.6 Pro. Rybin did not give the neural network any hints or specific mathematical instructions. Instead, he used only four general prompts, such as "make a breakthrough," "disprove the general case," and "continue exploring." The model independently constructed a logical chain and, within a few hours, provided a counterexample disproving the conjecture.

The researcher himself called this chat a "pure meme," emphasizing that he did not expect such a result from an AI working with general formulations. However, despite the impressive result, I must note that independent verification of this disproof has not yet been conducted. The mathematical community, and in particular specialists in flow theory, must confirm the correctness of the found counterexample.

Analytical commentary: This case demonstrates that modern AI models are capable not only of processing known data but also of generating new, non-trivial solutions in areas where traditional algorithms have long failed. If the result is confirmed, we will witness a new stage in the symbiosis of humans and artificial intelligence, where the latter acts not just as a tool but as a full-fledged participant in scientific discovery. For the crypto industry, where mathematical rigor underpins protocol security, this is a signal to reconsider methods for checking vulnerabilities.