Brief

Anthropic docs point to a narrower enterprise control model for Claude agents

Anthropic’s Claude API docs now describe self-hosted sandboxes for Claude Managed Agents, shifting tool execution from Anthropic-managed cloud sandboxes into infrastructure the user controls while leaving orchestration with Anthropic. The documented split gives the host control over the agent’s code environment, filesystem, spawned processes, and reachable network, but it does not make the setup fully customer-contained because tool inputs and outputs still flow to Anthropic’s control plane where Claude runs. ClaudeDevs also posted on June 12 that new guides were added for Blaxel AI, E2B, Google Cloud, Namespace Labs, and Superserve AI, but the extracted Anthropic documentation in the research pack did not verify those provider-specific guides or outbound-only networking claims. The evidence supports a shift in where agent tools can run, not a proven reduction in compliance friction, data-residency risk, or vendor lock-in.

AI · June 12, 2026

Sources & context

Evidence, open claims, and reader-facing editor notes.

Dek

Anthropic’s managed-agent sandbox guides point to a tighter enterprise bargain: keep tool execution, data residency, and isolation under customer control while model inference stays with Claude.

Synthesis angle

Investigate whether self-hosted agent sandboxes change enterprise AI incentives by separating model access from execution infrastructure, compliance controls, and vendor lock-in.

Source

Anthropic Launches Self-Hosted Sandboxes for Claude Managed Agents

Preferred source plan

Anthropic Claude platform documentation for managed agents and self-hosted sandboxes
Anthropic API documentation on tool execution, outbound-only communication, and orchestration boundaries
Google Cloud, E2B, Blaxel AI, Namespace Labs, and Superserve AI product pages or integration docs for Claude managed-agent sandboxes
Customer case material or engineering posts from Rogo and CodSpeed verifying deployment claims
Enterprise security documentation covering isolation, logging, network egress, and data residency controls

Reading the dossier...

Anthropic docs point to a narrower enterprise control model for Claude agents

AI · June 12, 2026