Race to regain AI model supremacy: Why Google shifted power in its top ranks

In recent months, Google co-founder Sergey Brin has encouraged top AI staff to focus on Google's Gemini model as the company seeks to catch up with rivals, according to two sources familiar with his remarks. Anthropic, on the other hand, has made significant progress with its powerful Claude Mythos model. Brin addressed over 200 employees in a town hall in April, urging DeepMind to speed up their work, a person with knowledge of the meeting said.

Gemini briefly surpassed competitors in November, but Google delayed the new version by two months after internal testing revealed it lagged behind rivals in areas such as coding. On August 5, Google announced a major leadership overhaul of its Google DeepMind AI division, with Demis Hassabis stepping down as chief and Koray Kavukcuoglu becoming his deputy. Hassabis became chair, with fewer management obligations. Two original technical co-leads of Gemini also quit to co-found a startup.

While some industry observers have questioned if Google is exiting the model race, interviews with seven knowledgeable sources suggest that the reorganization is tied to Alphabet's effort to achieve model supremacy. The changes mark a further erosion of DeepMind's autonomy, which Google has gradually undermined since acquiring the London-based lab in 2014.

During an all-hands meeting on August 6, it was revealed that some teams were moving out of DeepMind and into corporate Google. The move follows a 2025 Reuters special report that detailed how Hassabis had sometimes hindered efforts that could have generated new revenue for Alphabet or improved its standing in the AI race. Brin, who no longer holds an executive title, has been informally pushing staff to catch up to the frontier in AI, particularly in areas like recursive self-improvement, according to one source.