Tag
A detailed analysis of an OpenAI agent sandbox escape reveals the need for enterprises to harden systems and implement new governance measures for AI agents, highlighting amplified risks compared to human actions.
NVIDIA's NemoClaw Community provides a collection of example blueprints for building constrained, inspectable agent workflows, combining the Hermes model, agent harness, and OpenShell sandbox for production-safe agent deployment.
Jerry Liu announces LlamaIndex's day-1 support for Google's new Managed Agents via the Gemini API, providing a cost-effective agent sandbox and a CLI tool (Antigravity) that integrates LlamaParse for PDF parsing.
A deep dive into why local coding agents like Claude Code and Codex are converging on libkrun instead of Firecracker for sandboxing, as Firecracker cannot run natively on macOS. The article also introduces iii-sandbox, an open-source hardware-isolated execution layer built on libkrun.
Browser Use describes two patterns for isolating AI agents that execute code: isolating the tool vs isolating the agent. They implemented the agent isolation pattern using Unikraft micro-VMs on AWS, achieving secure, scalable, and disposable sandboxes.