Anthropic has launched web and mobile interfaces for its popular Claude Code AI coding tool, alongside a new sandboxing runtime designed to improve security and reduce user friction. The updates allow developers to manage tasks more efficiently while maintaining protections against risks like prompt injection. These features are now available in beta for Pro and Max subscribers.
Claude Code, Anthropic's command-line interface (CLI) agentic AI coding tool, has gained web and mobile access points. The web interface is fully developed at launch, enabling developers to connect it to a GitHub repository and issue broad instructions, such as “add real-time inventory tracking to the dashboard.” Like the CLI version, it provides progress updates and now supports mid-task suggestions or changes, avoiding the need to cancel and restart if issues arise.
Developers can run and switch between multiple sessions via a left-side panel. The mobile version is iOS-only and in early development.
More notably, Anthropic introduced a sandboxing runtime to enhance security and streamline interactions. Previously, Claude Code required permissions for most changes, leading to frequent approvals. Now, users grant access to specific file system folders and network servers, reducing interruptions while bolstering defenses against prompt injection and other threats.
As detailed in Anthropic’s engineering blog, internet access is restricted "through a unix domain socket connected to a proxy server running outside the sandbox. … This proxy server enforces restrictions on the domains that a process can connect to, and handles user confirmation for newly requested domains." Users can customize the proxy for their own traffic rules, allowing safe actions like fetching npm packages from approved sources without full external access or constant prompts.
These changes enable more independent operation for Claude Code agents, which many developers value over the new interfaces. However, the reduced oversight means code reviews are crucial, as the old approval system ensured close examination of changes, potentially making errors easier to overlook now.
The features are in beta as a research preview, accessible to Claude users with Pro or Max subscriptions.