What changed
Anthropic introduced self-hosted Managed Agents on November 12 in private beta. Beyond the earlier sandbox option, teams can operate the harness, workers, database, credentials and session history in their own cloud or on-premises. Model requests go to a selected provider or gateway.
One useful next step
Draw one proposed workflow with its input store, agent server, sandbox, model endpoint and logs. Identify what crosses each boundary, who can stop the run and how secrets stay separate. Use synthetic data; do not deploy simply to try the architecture.
Define an architecture review before a trial ↗Before you dive in
Access is private beta through an Anthropic account team. Network limits, credential isolation and audit records are documented design features, not an independent security assessment. Self-hosting also leaves an operator responsible for running the service.
Source & announcement date
Anthropic ·
Anthropic: Operate Claude Managed Agents inside your security perimeter — November 12, 2026 ↗ (opens in a new tab)This brief summarizes the linked announcement. “The AI Superpower take” and the next step are our editorial interpretation, not claims from the source.