On July 21, 2025, Demis Hassabis, DeepMind’s co-founder, logged onto Twitter. He was particularly interested in talking about the new changes to artificial intelligence and how it affects the world of competitive mathematics. His comments come at a time of buzzing conversations concerning the International Mathematical Olympiad (IMO). In these conversations, evaluators are looking at generative AI models and testing their mathematical prowess. Indeed, perhaps equally as remarkable has been the role that both OpenAI and Google have played and not played in the headlines surrounding this tumultuous competition.
OpenAI, the organization behind some of the most advanced natural language processing systems, chose not to compete formally in the IMO. Instead, the organization had no clue that it would earn a gold medal score in the very high range until those results were revealed. In the wake of this revelation, OpenAI contacted the IMO organizers to ascertain its status. The IMO has asked OpenAI to delay any public communication until after its Friday night awards ceremony.
Maxwell Zeff is a senior reporter at TechCrunch, where he covers AI. AI is transforming just about every sector, including education. At the same time, traditional education competitions are becoming more competitive. Zeff quoted Thang Luong, a senior researcher at Google DeepMind and lead for the IMO project. In particular, he stressed that the IMO organizers use their own internal grading rubric. So any evaluation that’s not founded on that guideline would not be able to claim gold-medal level performance.
Google is no stranger to the IMO’s podium, having won the silver medal in 2014. It earned this distinction by employing a detailed, objective grading rubric. The difference between formal participation and informal evaluation is dramatic. This discrepancy highlights the unique hurdles AI actors face when attempting to operate in the competitive space. In contrast to Google’s structured approach, OpenAI hired third-party evaluators to assess its model’s performance against the IMO standards.
Hassabis underlined the need for strict compliance with agreed protocols in academic circles. In doing, he reiterated on the decision to not immediately announce results as soon as they were verified by independent experts.
Btw as an aside, we didn’t announce on Friday because we respected the IMO Board’s original request that all AI labs share their results only after the official results had been verified by independent experts & the students had rightly received the acclamation they deserved – Demis Hassabis
Maxwell Zeff’s reporting is an excellent window into the quickly changing role that AI is playing in competitive spaces. His coverage always reminds readers of that oft-lost balance between technological advancement and long-held academic values. When he’s not chasing down AI reporting, Zeff enjoys hiking and biking. He loves the Bay Area food scene’s incredible diversity and creativity.
In short, unlike real-world applications of AI, this AI performance raises red flags about fairness and evaluation standards. AI is developing and maturing quickly, and AI is being widely adopted in the industry. Philanthropic organizations like OpenAI and government-supported ones like Google need to be sensitive to these shifts.
