Gemini surpasses one billion users as Hassabis proposes a joint safety body

Two stories from the same week, both concerning the Google galaxy, tell of the sector’s double speed: the spread of products races ahead, the building of common rules lags behind.

The billion

On August 11, 2026 Sundar Pichai announced that the Gemini application had surpassed one billion monthly active users. The company presents it as its fastest-growing product and as the fourteenth of the group’s services to reach that threshold, alongside Search, Gmail, Android and YouTube. According to data reported by TechCrunch and picked up in the AI Weekly roundup, 63% of users make use of the application’s voice features.

The story currently stems from a single source — the Alphabet leadership’s announcement, reported together with the TechCrunch figure in the cited roundup; no independent confirmation is available on the figures. This is a necessary clarification: usage figures are produced by the company that benefits from them, using counting criteria it defines itself. “Monthly active user” is a product metric, not a measure verified by third parties, and in an application that is preinstalled or built into already widespread services, the line between chosen use and encountered use is thin.

The figure on voice features, if confirmed, is the more interesting of the two: it indicates that for nearly two thirds of users, interaction with the model does not go through a keyboard. This is a shift that concerns accessibility — those who do not write easily can still take part — and one that changes the shape of the data collected, because voice contains information that text does not.

In the same period Meta released Muse Glimmer, a dense multimodal model with 30 billion parameters under an Apache 2.0 license and a context window of 131 thousand text units, able to run on a single consumer-grade graphics card: the opposite path to that of models accessible only through the data centers of those who own them.

Hassabis’s proposal

On the rules side, Demis Hassabis, head of Google DeepMind, has discussed the creation of an independent industry body tasked with codifying common safety practices for the development of advanced artificial intelligence. According to reporting by the Wall Street Journal and picked up by Tech Startups, the proposal was put to executives from the leading labs in the sector and to US officials, including Treasury Secretary Scott Bessent and White House technology adviser Michael Kratsios. Hassabis reportedly drew a conceptual comparison with institutions such as the International Atomic Energy Agency.

Here too the story currently stems from a single source — the Wall Street Journal article, picked up in the cited roundup; no independent confirmation is available.

The comparison with the atomic agency must be handled with care, because that body arises from a treaty between states and holds inspection powers, whereas an industry body is by definition a structure that the regulated parties set up for themselves. These are two models bearing the same name and two opposite sources of legitimacy: in the first case authority comes from member states, in the second from the will of the companies that take part in it and that can leave it.

Why now

The proposal is taking shape in a specific context: frontier labs are under competitive pressure, and governments are struggling to build regulatory frameworks that keep pace with the capabilities of the models. In Europe the public response is already operational — since August 2 the Commission’s AI Office has been applying the transparency rules of the AI Act, as reported today in the Europe section. In the United States, where no equivalent framework exists, the idea circulating is that of a codified self-regulation.

The two stories are of a different order of magnitude: one billion users is an accomplished fact measured by the company, a joint safety body is a conversation among executives and officials with no founding act.