AI Runtime

NVIDIA has published OpenShell, an open-source runtime designed to let fleets of autonomous AI agents run safely and privately. The project, hosted on GitHub, is drawing significant attention from developers discussing sandboxing, isolation and security for large-scale agent deployments, as interest in production-grade infrastructure for multi-agent systems accelerates.

Developers running servers, scrapers or bots on Node.js are being urged to update tonight after a bug in Google's Brotli decoder, tracked as CVE-2026-111497, was disclosed. Decoding crafted Brotli data can crash a process, leak memory or allow remote code execution. Node.js has shipped patches in versions 22.23.4 and 24.22.1.

NVIDIA has published OpenShell, an open-source runtime written in Rust described as a safe, private environment for running autonomous AI agents. The project is drawing attention on GitHub, where it has quickly climbed the trending rankings as developers examine its sandboxing and privacy approach for agent workloads.

A new open-source project has been released that provides persistent memory for AI agents across different harnesses, meaning agents can retain and share context regardless of which framework or runtime is running them. Discussion so far focuses on the technical appeal of making agent memory portable and interoperable rather than locked to a single tool, with early readers weighing in on the design approach.

A new blog post describes a method for detecting tampering with AI model weights directly in GPU memory, focusing on so-called 'Rome attacks' that modify model behaviour at inference time. The author walks through how such tampering can be spotted in practice. The topic is drawing attention from machine learning engineers concerned with AI security and supply-chain integrity.

Stability AI has published Orca, an open-source TypeScript project for managing a fleet of AI coding agents such as Grok Bot and Claude Code. The tool lets developers run any coding agent with their own subscription, across desktop, mobile and a remote runtime, and is already drawing attention from developers on GitHub.

A monthly ranking of the fastest-growing Rust projects on GitHub shows stars flowing to agent browsers, sandbox runtimes, memory layers, GUI frameworks and graph databases. The list suggests developer attention in the Rust ecosystem is concentrated around AI agent infrastructure and supporting tooling, alongside long-standing interest in graphical and data-layer projects.

Anthropic has released sandbox-runtime (srt), the open-source, operating-system-level sandboxing technology behind Claude Code. The tool restricts file and network access for AI agents, MCP servers and command-line tools, using Seatbelt on macOS, without requiring containers. Developers see it as a practical way to run AI agents with less risk of unintended system changes.

Amazon Web Services has introduced Infinity, a runtime designed to handle extreme agent concurrency, allowing large numbers of AI agents to run in parallel. The announcement highlights AWS's push into infrastructure for agentic AI workloads. Developers are watching to see how it performs against existing orchestration tools and what it means for scaling multi-agent systems.

IBM is introducing identity and runtime controls aimed at reining in rogue AI agents, addressing the growing problem of autonomous software acting beyond its intended scope. The move reflects a broader push in the tech industry to bring security and accountability frameworks to agentic AI systems as companies increasingly deploy them in production environments.

Google chief executive Sundar Pichai says Search is being absorbed into an agentic layer, with Gemini and Search sharing a single model, harness and runtime for consumers and enterprises alike. He argues query-driven discovery is giving way to task-driven assistants, a shift with major implications for how brands and websites remain visible in Google's ecosystem.