The most important AI stories, curated from top research labs, tech outlets, and industry newsletters — in one 3-minute read.
One email per week. No spam, ever. Unsubscribe anytime.
This week:
– Model race heats up: major upgrades across Gemini, Claude, OpenAI
– Strategic AI infrastructure expands in US, India, Meta/NVIDIA
– Corporate shake-ups and regulatory tests for big tech
– AI safety, security, and provenance push forward
Google released Gemini 3.1 Pro, its most advanced model yet for complex, multi-step tasks, boasting double the ARC-AGI 2 performance of Gemini 3.0, a massive 1M token context window, and a clear push toward enterprise dominance, as reflected across broad coverage, industry analysis, and benchmarks.
Anthropic launched Claude Sonnet 4.6, claiming Opus 4.5-level performance in its mid-tier model and better tool use—analysts note mostly clean gains with a few caveats as Claude tightens mid-size competition.
It emerged that OpenAI detected and banned a future Canadian school shooter for violating ChatGPT’s usage rules eight months prior, but ultimately did not refer her to police, fueling debate on AI platforms’ duty to warn.
Microsoft’s Xbox and Gaming CEO Phil Spencer and president Sarah Bond announced their departures, with AI executive Asha Sharma taking over and the company reaffirming commitment to consoles.
Meta and NVIDIA expanded their alliance, as Meta becomes one of the first hyperscalers to deploy NVIDIA CPUs at scale alongside millions of GPUs, signaling a new era in data center AI infrastructure.
Industry & infrastructure
Google added AI music generation (Lyria 3) to Gemini, letting users create tracks from text and images.
India’s telcos and conglomerates announced over $200B in AI datacenter investments, with PM Modi positioning the move as job-positive.
OpenAI sealed a data center deal and Cerebras plans an AI supercomputer for India.
GitHub launched Agentic Workflows in technical preview for AI-driven repo automation.
GGML and llama.cpp joined Hugging Face to support the long-term future of local AI models.
NVIDIA released new enterprise AI tools, including DreamDojo, an open-source robot world model trained on 44,711 hours of human video, and Dynamo v0.9.0, a major distributed inference update.
Meta shifted Horizon Worlds to focus almost exclusively on mobile, decoupling it from the Quest VR platform.
AI safety, security & policy
Anthropic debuted Claude Code Security, prompting cybersecurity stocks to drop as the tool aims to automate AI code review.
Tesla lost a bid to overturn a $243M Autopilot verdict and complied with a California DMV order to drop "Autopilot" branding in marketing.
Microsoft outlined strategies for media authenticity and provenance, as deepfakes proliferate and industry scrutiny grows.
Wikipedia banned Archive.today, removing 695,000 archive links over DDoS and tampering allegations.
NASA’s Artemis II mission faced hardware setbacks, forcing a moon mission delay as repairs require taking the rocket back to its hangar.
Agentic tools & methods
OpenAI introduced Harness Engineering, letting Codex agents build, test, and deploy million-line software systems.
LangChain upgraded its Agent Builder with enhanced chat, file uploads, and a tool registry, while also detailing the memory system behind agent workflows.
Anthropic analyzed agent autonomy, finding AI agents thrive in software development, but usage elsewhere remains rare.
Funding & market activity
AI startups dominated recent funding rounds, with World Labs landing $1B for 3D spatial intelligence and over $9B poured into seed rounds across robotics, multimedia, and backend automation.
Until next week.