Skip to content
AISecurity

Microsoft says Windows 11 is an agentic platform—and only OS sandboxing can stop rogue agents

theregister / os-platforms 12h ago 7

At Microsoft Build 2026, Satya Nadella positioned Windows 11 as an agentic platform, arguing that AI agents are just another user class that operating systems must manage and secure. A live demo showed OpenClaw trying to delete desktop files while locked in a sandboxed Windows MXC container with granular permissions, proving that trust and control are prerequisites for agent acceptance. The author argues that while keynote visions hype effortless omniscient agents, Microsoft's rediscovery of core OS security principles offers a more realistic, if currently cumbersome, path to safe agentic AI.

Read full article →

More AI

Biohub open-sources AI world model for protein biology and drug design

hackernews / top 1d ago 9

Biohub releases open AI models that map the protein universe and design lab-validated therapeutic binders in days instead of years.

Google to pay SpaceX $920M monthly for xAI data center GPU capacity

hackernews / top 1d ago 9

Google will pay SpaceX $920 million a month for 32 months to rent roughly 110,000 GPUs inside xAI data centers, giving Google Cloud bridge capacity for Gemini Enterprise as SpaceX prepares its record IPO.

OpenAI Ships Million-Line Product Written Entirely by Codex Agents

hackernews / top 1d ago 9

OpenAI details building and shipping a full product with zero manually-written code using Codex agents, achieving roughly 10x speed and redefining engineering as steering rather than coding.