Anthropic Announces Self-Hosted Claude Managed Agents for Enterprises

On November 12, 2026, Anthropic announced a private beta of self-hosted Claude Managed Agents. The new offering lets enterprises run Managed Agents servers, credentials, sandboxes, and session history in their own cloud environment or on-premises, rather than in Anthropic's infrastructure.

The announcement addresses one of the biggest barriers to enterprise adoption of AI agents: control. Large companies, banks, hospitals, law firms, and government agencies often cannot send sensitive data, credentials, or internal work to an outside provider's servers. Self-hosting removes that barrier by keeping the agent infrastructure inside the customer's own environment.

What Claude Managed Agents Are

Managed Agents are Anthropic's managed service for running Claude agents — AI systems that can work through multi-step tasks, use tools, access data, and carry out work on behalf of a user, rather than simply answering a single question.

In a standard managed service, the provider runs the infrastructure: the servers that host the agents, the systems that store credentials, the sandboxes where code runs, and the history of past sessions. That is convenient, but it means the customer's data and credentials live outside its own walls.

The self-hosted version changes that arrangement. Enterprises run the Managed Agents servers themselves, in their own cloud account or in their own data centers. They hold the credentials agents use to access internal systems, control the sandboxes — the isolated environments where agents run code and tools safely — and retain session history, the record of what agents did, inside their own environment.

No pricing, availability date for general release, or list of supported cloud providers was included in the sources for this article. As a private beta, access is limited, and details of the rollout are undisclosed.

Why Self-Hosting Matters

Self-hosting is not a technical detail. For many enterprises, it is the difference between being able to use AI agents and not being able to use them at all. Each of the four components Anthropic names carries a different risk if it leaves the enterprise.

Servers determine where the agent runs and where data is processed. Credentials are the keys, tokens, and passwords that let an agent access email, databases, internal tools, and customer systems; if those are stored outside the enterprise, a breach could expose internal systems. Sandboxes are where agents execute code — if code runs outside the enterprise, the enterprise loses control over what it can touch. Session history is the record of agent activity, which can itself contain sensitive information that many companies are required by law or contract to keep inside their own systems.

By letting enterprises host all four components themselves, Anthropic is making a direct pitch to regulated and security-conscious organizations — the customers with the largest budgets and the strictest requirements, and often the slowest to adopt AI agents precisely because of these control issues.

The Enterprise AI Agent Market in 2026

The announcement fits a broader pattern in 2026. AI agents have moved from demos to real deployments, and the conversation has shifted from what agents can do to how they can be deployed safely inside large organizations.

Three deployment models now dominate the enterprise conversation. The first is fully managed, cloud-based agents, where the provider runs everything. The second is API-based agents, where the enterprise builds its own agent infrastructure on top of a model provider's API. The third is self-hosted or private-cloud agents, where the model and agent infrastructure run inside the customer's environment.

Each has trade-offs. Fully managed services are fastest to start but offer the least control. API-based agents offer flexibility but require the enterprise to build and maintain its own infrastructure. Self-hosted managed agents aim for a middle ground: the provider supplies the agent software and management, while the customer supplies the infrastructure and retains control.

Why It Matters for Founders and Investors

For founders building AI agent products, the announcement signals what enterprise customers will demand. If you sell agents to large companies, expect questions about where data lives, where credentials are stored, where code runs, and where history is kept. A product that cannot answer those questions may not pass security review, no matter how good its agent is. Founders should design for deployment flexibility from the start.

For founders in regulated industries — finance, healthcare, legal, insurance — self-hosted agents could unlock use cases that were previously blocked. A law firm could run agents on client matters without sending client data outside the firm; a bank could run agents on internal systems without exposing credentials to a third party. Those use cases are high-value, and enterprises are often willing to pay most for them.

For investors, the announcement shows Anthropic competing directly for the enterprise agent market, not just for developers or consumers. Enterprise contracts are larger, longer, and more stable than consumer subscriptions, and they are the revenue base that justifies the large valuations of frontier AI companies. The private beta status is a reminder that enterprise AI is still early — the products that win will solve deployment and control problems, not just model capability.

Angeled.com connects founders with investors.