Google DeepMind CEO pushes for AI watchdog in Washington

Summary

CNBC's MacKenzie Sigalos reports on Google DeepMind CEO Demis Hassabis' push for a permanent AI regulatory body, organizational bottlenecks slowing Google's AI launches, rising use of Chinese AI models by U.S. companies, and a short-lived bump in Google shares on chip efficiency news.

• DeepMind CEO Demis Hassabis urges Congress to create a permanent, FINRA-style AI watchdog with pre-release model testing.
• Head of Trump administration's AI security agency resigns after three months, per Axios.
• Google proposes an industry-funded, federally overseen body that could eventually require model approval for U.S. entry.
• DeepMind faces delays in releasing Gemini Pro; senior researchers have left, while Chinese models from Moonshot AI and Alibaba narrow the capability gap.
• 45% of token use among U.S. companies now comes from Chinese open-source models, with no current restrictions on use.
• Google shares rose on a report of a specialized AI chip that could run Gemini 10x more efficiently, but organizational bottlenecks are the bigger problem.
• Multiple approval layers and too many decision makers slow DeepMind's launches, leaving Google technically competitive but often beaten to market.