OpenAI, Google, Microsoft, Anthropic launch major AI upgrades

This week saw a flurry of AI releases from major players. OpenAI launched Dreaming V3, a memory system for ChatGPT that integrates past chat details into user profiles, doubling storage for Plus/Pro users and adding Temporary Chats for privacy. Google DeepMind released Gemma 4 12B, a 12-billion-parameter multimodal model that runs entirely on devices with 16 GB RAM, supporting text, image, audio, and video offline with a 256K context window. Microsoft unveiled Scout, an always-on AI assistant embedded across Teams, Outlook, and Windows, built on OpenClaw for agent-first workflows, alongside Project Solara, an Android-based OS for AI-driven devices. Anthropic launched Claude Opus 4.8, emphasizing honesty with dynamic workflows for up to 1,000 subagents, faster modes, and improved code flaw detection.

Beyond product launches, trends in privacy and security shaped the narrative. DuckDuckGo saw a 70% surge in US iPhone installs and record traffic after backlash to Google's AI search results, launching Chrome/Firefox extensions for AI-free search. Microsoft introduced Execution Containers (MXC) at Build 2026, an OS-level sandbox restricting AI agent access. Nvidia and Microsoft jointly announced the RTX Spark superchip (Arm-based, 20-core Grace CPU, Blackwell RTX GPU, 128 GB unified memory) for on-device AI reaching one petaflop. Pinterest signed a $4B cloud/AI deal with AWS using Trainium chips. Zoom introduced ZoomMate, turning meetings into actionable tasks across Salesforce, Jira, and Slack. The message: AI is moving closer to devices and workflows, but security and privacy concerns are equally accelerating.