article --summary
The summary
Anthropic has opened a private beta of a self-hosted version of its Claude Managed Agents platform, letting customers run both the coordinating server and the sandboxes where agents execute inside their own cloud or data center. The reader will learn how this differs from the earlier hosted and partly self-hosted options and why the company's architecture keeps credentials and session history on customer systems.
The main concern raised is the license. If Anthropic flags a safety, security, or misuse risk in writing, customers have eight hours to disable the feature or roll back across every deployment, and one business day to install a required safety update. The license also prevents customers from modifying the proprietary server, restricts them to Claude models unless Anthropic approves others, lets Anthropic reserve certain models for its hosted service, and bars them from blocking the usage telemetry sent back to the company.
Analysts quoted in the piece expect regulated buyers such as banks and government agencies to push back on the rollback deadline and telemetry terms, while suggesting that buyers needing full control over their agents may look to independent or open-weight alternatives instead.
Editorial note: This is a summary of an article first published by The New Stack. The full piece is on their site.