Google and DeepMind launched the DeepMind Institute to advance AGI research. The board features Demis Hassabis and Shane Legg. Initial articles cover AGI economics, reasoning transparency, and evaluation frameworks. Researchers warn about shrinking model transparency, suggesting limits on invisible serial depth. Hassabis proposed a US-led frontier AI standards body for voluntary pre-release assessments, potentially becoming mandatory. The institute aims to shift focus from capability development to governance, transparency, and independent confidential testing to ensure safety and controllability of advanced AI systems.