🤖 Gemini Robotics ER 1.6 gives robots spatial reasoning 🤖
Google DeepMind's latest robotics model bridges language reasoning with physical-world precision.

Google DeepMind has released Gemini Robotics-ER 1.6, a major upgrade to its embodied reasoning model that gives robots sharper spatial awareness and multi-view understanding. The standout new feature is instrument reading — the ability to interpret complex gauges, sight glasses, and digital readouts — developed in close collaboration with Boston Dynamics for autonomous facility inspections using Spot. The model also improves pointing accuracy, task planning, and success detection, letting robots know when a task is finished and decide whether to retry or move on. This closes the gap between digital intelligence and physical action. Available now via the Gemini API and Google AI Studio, it is also Google's safest robotics model yet, with improved compliance on adversarial safety tasks.

🛡️ GPT-5.4-Cyber counters Anthropic's Mythos with open access 🛡️
OpenAI opens broad cybersecurity AI access where Anthropic kept Mythos restricted.

OpenAI has released GPT-5.4-Cyber, a purpose-built variant of its flagship model designed exclusively for defensive cybersecurity operations. Distributed through the Trusted Access for Cyber initiative, the model is available to thousands of verified users who pass identity checks — a notably broader access strategy than Anthropic's Mythos, which remains limited to a small group of partners. GPT-5.4-Cyber can reverse-engineer compiled software to detect malware and vulnerabilities without requiring source code, significantly accelerating analyst workflows. The contrasting deployment approaches are shaping how frontier AI reaches cybersecurity teams. The release positions OpenAI as the early front-runner in open-access AI for threat detection.

⚡ Claude Code adds scheduled Routines via GitHub ⚡
Anthropic's GitHub integration turns Claude Code into a scheduled automation layer.

Anthropic has launched routines in Claude Code, a research preview feature that lets AI tasks run on a schedule, via API, or triggered by GitHub events. Routines come with their own dedicated API endpoint and can subscribe to GitHub webhooks, enabling automated developer workflows without manual intervention. Available on all paid plans with Claude Code on the web enabled, the feature extends Claude Code from an interactive coding assistant into a full task-execution engine — a key step toward owning the developer automation layer. By targeting CI/CD and DevOps workflows where event-driven automation is critical, Anthropic is making a clear play for the infrastructure backbone of AI-assisted software development.

INTERESTING TO KNOW

🖥️ Google builds a desktop agent for Gemini 🖥️

Google is launching a new desktop agent system for Gemini that moves beyond chat-based prompts to full autonomous task execution across apps like Gmail, Drive, Calendar, and the web. The update introduces a dedicated Agent tab with a structured workspace for managing goals, tools, files, and connected services, plus a "require human review" toggle for approving actions before execution. The system handles multi-step workflows such as booking trips and clearing inboxes, directly pressuring OpenAI and Anthropic in the enterprise agent race.

📧 Microsoft Copilot now handles email delegation 📧

Microsoft has introduced email delegation for Copilot, letting users forward emails directly to the AI assistant for autonomous handling. This integration moves Copilot deeper into agentic territory — taking action on behalf of users rather than just responding to prompts. The feature targets enterprise inbox management and positions Microsoft competitively against Google's Gemini desktop agent and Anthropic's expanding tool capabilities, marking a concrete step toward fully agentic workplace assistants across the Office ecosystem.