On October 6, 2026, US-based Anthropic discontinued the local execution option for its “Claude Cowork” AI agent feature, completely migrating its processing infrastructure to the cloud. Behind this decision are factors such as the device load from local virtual machines, issues associated with OS updates, and security concerns regarding sandbox escapes.

Overview of the Discontinuation of Local Execution Options and Complete Cloud Migration

On October 6, 2026 (local time), US-based Anthropic discontinued the “This computer only” option, which executes tasks on local PCs, for “Claude Cowork,” an AI agent feature previously offered to paid Claude plans Pro and Max. Tasks that were previously processed in virtual machines (VMs) on the user’s local PC will now be entirely centralized in Anthropic’s cloud environment, including both model inference and the execution VMs. While local tasks already initiated prior to October 6 will continue to run until completion under their initial settings, the specification has been changed so that any new tasks created on or after that date will be processed in the cloud environment without exception. Claude Cowork is a powerful feature that autonomously executes multi-step operations such as file manipulation and tool integration to create deliverables simply by being given objectives and instructions. Due to this change, the mode switching that allowed users to consciously select between local execution and cloud execution has been eliminated, integrating the feature into a simpler user experience. Please also review the system configuration diagram below.

Technical Background Leading to Discontinuation and Security Vulnerabilities

Behind the discontinuation of the local execution option lie technical operational failures and critical security vulnerabilities. The Windows 11 security update “KB5124008,” distributed on September 8, 2026, caused a massive failure that broke folder sharing between the host PC and Linux VMs via the Hyper-V infrastructure and the Plan9 sharing mechanism, making local command execution impossible. Furthermore, the security research organization Accomplish AI pointed out a sandbox escape vulnerability dubbed “SharedRoot” (related to CVE-2026-46331) in the macOS version of Claude Cowork, which allowed unauthorized access from within the local VM to the entire root directory (/) of the host OS. If exploited, this vulnerability posed a risk of sensitive files, such as SSH keys and cloud credentials, being read from and written to externally. Additionally, operational challenges—such as running local VMs placing an excessive load on the user’s PC resources and work being interrupted when closing a laptop lid—have been major factors driving the shift to the cloud.

Changes in Technical Architecture and Operations Due to Cloud Migration

Secure Data Synchronization Mechanism via Desktop App

Although the execution mechanism has been changed so that tasks are run in the cloud, the integration feature with folders and files on your local PC is still retained. When Claude accesses local files, the architecture routes through the desktop application (Claude Desktop) installed on the PC, restricting access only to specific folders that the user has explicitly permitted to connect. If files become necessary during cloud-based work processes, only duplicates (copies) of the required data are sent to a dedicated sandbox VM on the cloud for processing. Each session is assigned an independent sandbox environment that does not share state with others, featuring a secure design where duplicated data on the cloud is immediately erased as soon as work records are deleted or the session is ended. On the user side, it is recommended to manage operations by narrowing down access permissions to a minimum to prevent folders containing confidential information from being accidentally included in the connection targets. An image illustrating the synchronization mechanism is as follows.

The Convenience and Changes to the Work Environment Brought by Cloud Execution

With the complete migration of the processing infrastructure to the cloud, user convenience and operational efficiency have significantly improved. The biggest advantage is the ability to execute large-scale document analysis and complex task processing without consuming the hardware resources (such as CPU and memory) of your local PC. By eliminating the computational load on the device, even if you close your laptop lid or turn off the PC, AI agents can now continue working in the background on the cloud. Furthermore, multi-device integration has become seamless, making it easy to check the progress of long-running tasks started in a desktop environment or review generated outputs from mobile devices such as smartphones and tablets while on the go. Regularly scheduled tasks, such as daily or weekly routines, are now reliably executed in the cloud without having to worry about whether your local PC is powered on, substantially enhancing the reliability of business automation.

Operational Measures Required of Users and Future Outlook

Alternatives for Local Processing and the Use of Claude Code

For cases where users handle highly sensitive data and want to complete processing entirely on their local device without sending any information to an external cloud, or for users who want to perform advanced local command operations, alternatives are available. Anthropic recommends the use of “Claude Code” as a tool that runs directly in the local PC environment and keeps files and history locally. Although Claude Code originally emerged as a command-line tool for developers, its integration into desktop apps and other platforms is progressing, making it possible to perform file operations and execute tasks in the local environment directly from the terminal or app. Furthermore, by installing third-party extensions such as Desktop Commander MCP, the option to individually build and maintain a local machine control environment is also being utilized within the community. It is important to clearly choose between the cloud-based Cowork and the local-based Claude Code according to your specific use case.

Future Developments and Key Points for the Safe Use of AI Agents

The recent specification changes to Claude Cowork can be described as a crucial turning point for safely and stably integrating advanced AI agents into society. While the convenience of allowing AI to directly operate local PCs offered clear benefits, it also brought to light risks such as OS compatibility issues and sandbox security vulnerabilities. As a result, secure cloud-based sandbox isolation and access control are expected to become the industry standard moving forward. Enterprise users and individual users alike will be increasingly required to implement strict permission management—such as restricting connected folder ranges to the absolute minimum—and establish governance structures where humans conduct final reviews of the numerical data and factual accuracy in generated outputs. Going forward, the division of roles between the security governance of cloud-based autonomous AI agents and developer tools operating in local environments is expected to become much more distinct, strongly fostering the sophistication of safe and secure AI adoption.

Reference Page

• [Claude Cowork Support Page] https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile#h_f951c27c48
• [Claude Status Official Page] https://status.claude.com/
• [Microsoft KB5124008 Support Information] https://support.microsoft.com/ja-jp/servicing/os/windows-11/2026/09/kb5124008-windows-11-24h2-25h2-security-update
• [Microsoft Windows 11 Known Issues] https://learn.microsoft.com/hu-hu/windows/release-health/status-windows-11-24H2
• [Claude Cowork Official Product Page] https://claude.com/product/cowork

【#Claude #ClaudeCowork #AIAgents #Anthropic #CloudMigration #Security #ClaudeCode】