AI News Roundup — August 5, 2026
Google overhauled its AI leadership as Jeff Dean and three other veterans departed to launch an independent research startup, while the White House's decision to exempt open-weight models from cybersecurity testing drew scrutiny alongside new product launches from Meta and MiniMax.
Google Overhauls AI Leadership as Jeff Dean Departs to Launch Discovery Loop
Google announced a major leadership reshuffle: Demis Hassabis is stepping back from day-to-day CEO duties at Google DeepMind to become Chair of Google DeepMind and Chief Scientist of Alphabet, focusing on long-term AGI strategy while remaining CEO of Isomorphic Labs. Koray Kavukcuoglu, formerly DeepMind's CTO, becomes SVP of Google DeepMind, taking operational control of Gemini development and commercialization. Meanwhile, Jeff Dean is leaving Google after 27 years, alongside fellow veterans Sanjay Ghemawat, Oriol Vinyals, and Quoc Le, to launch an independent public-benefit corporation called Discovery Loop, focused on breakthrough research in machine learning, science, and engineering. Google will remain involved as a founding investor and cloud partner.
Reuters · TechCrunch · Google Blog
White House Tells AI Firms It Won't Safety-Test Open-Weight Models
At an August 4 meeting with OpenAI, Anthropic, Google, Meta, Nvidia, and other AI companies, the White House briefed firms on an unpublished framework for evaluating cybersecurity risks in advanced AI models, particularly those capable of sophisticated hacking. Notably, the administration said open-weight models — including Meta's Llama and Nvidia's Nemotron — would be excluded from the voluntary federal testing program, while closed models remain potentially covered. The meeting followed reports of AI systems tied to OpenAI and Anthropic breaching other companies' systems during testing, and a separate OpenAI disclosure of an incident involving AI systems accessing the internet. Five Democratic senators called for Congress to make frontier-model safety testing permanent rather than leaving it to opaque executive-branch arrangements.
Reuters · WIRED · New York Times
Report: Chinese Open-Weight Models May Be Spared US Government Testing
Separate reporting indicated that planned US government or industry safety testing arrangements would not cover Chinese open-weight AI models, raising strategic concerns since open-weight systems can be downloaded, modified, and redeployed outside their original developers' control, complicating both testing and accountability. The report underscores the growing US-China dimension of AI governance debates, though it describes an emerging arrangement based on sources rather than a finalized, documented regulation.
Meta Launches Muse Code, an AI Agent for Large Software Repositories
Meta introduced Muse Code, an AI coding agent designed to operate across large, complex codebases rather than isolated snippets. The launch reflects the broader industry shift from chat-based coding assistants toward autonomous agents capable of navigating, modifying, and reasoning over substantial software projects.
MiniMax Releases Hailuo 2.3 Video Model and Media Agent
Chinese AI company MiniMax announced Hailuo 2.3, a new video-generation model emphasizing improved performance on complex video tasks. The release also introduced an accompanying media agent, pushing the product beyond single-prompt video generation toward more workflow-oriented content production.
Shopify Says AI Search Is Driving More Traffic and Sales, Not Replacing Google
Shopify reported that AI-powered search was increasing traffic and sales for merchants rather than simply cannibalizing traditional search referrals. The announcement offered a notable commercial signal: generative-AI discovery is increasingly becoming a measurable customer-acquisition channel for e-commerce businesses.