Anthropic details Linux container for Claude Cowork AI assistant

Anthropic has revealed the Linux container environment supporting its Claude AI assistant's Cowork mode, emphasizing security and efficiency. The setup, documented by engineer Simon Willison, uses ARM64 hardware and Ubuntu for isolated operations. This configuration enables safe file handling and task execution in a sandboxed space.

Anthropic's Cowork mode, an AI assistant for daily tasks and file operations, operates within a specialized Linux container environment. Technical staff member Simon Willison shared configuration details on GitHub Gist on January 25, 2026, highlighting the setup's balance of functionality and security.

The hardware features an ARM64 processor with four cores at 48 BogoMIPS, including advanced security elements like hardware cryptographic acceleration for AES, SHA1, SHA2, SHA3, and SHA512, atomic operations, pointer authentication (PACA/PACG), and branch target identification (BTI). Memory is limited to 3.8 GiB total, with 2.8 GiB available, and no swap space. Storage includes 10 GB NVMe for both root and session disks, mounted as ext4 filesystems, plus a 98 MB vfat EFI partition for /boot/efi.

Running Ubuntu 22.04.5 LTS (Jammy Jellyfish) on kernel 6.8.0-90-generic with PREEMPT_DYNAMIC for real-time support, the system uses aarch64 architecture and hostname 'Claude'. Sandboxing employs Bubblewrap (bwrap) for isolation, with independent network (--unshare-net) and PID (--unshare-pid) namespaces, die-with-parent termination, and measures against device hijacking.

Security is bolstered by Seccomp filter mode 2, NoNewPrivs to prevent escalation, and dropping all capabilities (CapEff = 0). Networking routes through local proxies: HTTP/HTTPS on port 3128 via UNIX sockets, and SOCKS5 on port 1080 for FTP/gRPC, using socat for tunneling to the host.

The execution stack starts with bwrap (PID 1), bash (PID 2), socat proxies (PIDs 3 and 4), and the main Claude process (PID 5) under model claud-opus-4-5-20251101. It supports tools like Task, Bash, Glob, Grep, Read, Edit, and Write, plus two Model Context Protocol (MCP) servers.

A related announcement for Claude Cowork v2 emphasizes container deployment for developers, using a Dockerfile from anthropic/claude-code base image, folder mounts to /sessions//mnt/ for granular permissions, and at least 4 GB RAM with 2 CPU cores. This aids cross-platform testing on Linux and Windows, reducing macOS dependencies and enhancing CI/CD integration while minimizing attack surfaces through namespace isolation.

Liittyvät artikkelit

Illustration of Claude AI controlling a Mac desktop, with open apps like Slack and Calendar, highlighting new research preview features.
AI:n luoma kuva

Anthropic's Claude AI Gains Full MacOS Desktop Control in Research Preview

Raportoinut AI AI:n luoma kuva

Building on its January Cowork feature, Anthropic has launched a research preview for Claude Code and Cowork tools, enabling Pro and Max subscribers' Claude AI to directly control Mac desktops—pointing, clicking, scrolling, and navigating screens for tasks like opening files, using browsers, developer tools, and app interactions such as Google Calendar and Slack. Safeguards address security risks, amid competition from tools like OpenClaw.

Anthropic has introduced Cowork, a new tool that extends its Claude AI to handle general office tasks by accessing user folders on Mac computers. Designed for non-developers, it allows plain-language instructions to organize files, create reports, and more. The feature is available as a research preview for Claude Max subscribers.

Raportoinut AI

Anthropic has confirmed the leak of more than 512,000 lines of source code for its Claude Code tool. The disclosure reveals disabled features hinting at future developments, including a persistent background agent called Kairos. Observers examining the code also found references to stealth modes and a virtual assistant named Buddy.

Anthropic is temporarily doubling usage limits for its Claude AI chatbot during off-peak hours from March 13 to March 27. The promotion applies to Free, Pro, Max, and Team plan users, excluding Enterprise plans. It activates automatically across web, desktop, mobile, and integrated apps.

Raportoinut AI Faktatarkistettu

After Anthropic CEO Dario Amodei said in late February that the company would not allow its Claude model to be used for mass domestic surveillance or fully autonomous weapons, senior Pentagon officials said they have no intention of using AI for domestic surveillance and insist that private firms cannot set binding limits on how the U.S. military employs AI tools.

US Defense Secretary Pete Hegseth has threatened Anthropic with severe penalties unless the company grants the military unrestricted access to its Claude AI model. The ultimatum came during a meeting with CEO Dario Amodei in Washington on Tuesday, coinciding with Anthropic's announcement to relax its Responsible Scaling Policy. The changes shift from strict safety tripwires to more flexible risk assessments amid competitive pressures.

Raportoinut AI

AI coding agents from companies like OpenAI, Anthropic, and Google enable extended work on software projects, including writing apps and fixing bugs under human oversight. These tools rely on large language models but face challenges like limited context processing and high computational costs. Understanding their mechanics helps developers decide when to deploy them effectively.

 

 

 

Tämä verkkosivusto käyttää evästeitä

Käytämme evästeitä analyysiä varten parantaaksemme sivustoamme. Lue tietosuojakäytäntömme tietosuojakäytäntö lisätietoja varten.
Hylkää