Google’s AI leadership comes apart in a single morning

Fast Company's AI Decoded newsletter reports significant upheaval at Google's AI department, with CEO Demis Hassabis and chief scientist Jeff Dean both stepping down on the same day. This move came in a memo from CEO Sundar Pichai, causing Alphabet shares to drop nearly 4%. Hassabis will assume the roles of chairman for Google DeepMind and chief scientist for Alphabet, continuing his leadership at Isomorphic Labs.

Meanwhile, Dean will co-found Discovery Loop, a public benefit corporation focused on automating the research loop for machine learning and engineering. Dean leaves alongside several other key figures - Sanjay Ghemawat, Oriol Vinyals, and Quoc Le - to spearhead Discovery Loop, which aims to streamline the research process before expanding to hardware design, drug discovery, and clean energy.

The article notes that this isn't the first major shift in Google's AI team, with other high-profile departures including Noam Shazeer to OpenAI and John Jumper to Anthropic. Ghemawat, speaking to The Wall Street Journal, cited Google's infrastructure limitations for scientific research, stating that the company's systems work well for consumer apps, ads, and search, but have "very different requirements" for scientific endeavors.

In other AI news, White House officials met with representatives from major AI companies, including Anthropic, OpenAI, Microsoft, Meta, Google, and Nvidia. The government announced they would review closed models for security risks, but not open-source models. This comes after the U.K. AI Security Institute logged nearly 20 hacking attempts by Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol during safety testing, though no malicious code escaped their test environment.

Finally, Meta has announced the launch of Muse Code, its first AI coding agent, as part of its Superintelligence Labs initiative. The new tool, running on Muse Spark 1.2, allows users to input natural language instructions and will plan, write, test, and verify code, potentially with multiple agents working in parallel.