Anthropic opened a self-hosted option for Claude Managed Agents
Anthropic published a product update dated November 12, 2026, introducing a self-hosted option for Claude Managed Agents. The official page gives that calendar date and no clock time. Help Net Security reported the release at 1:07 PM ET the same day (18:07 UTC), and Anthropic's product documentation describes the same option. This is a company product announcement, not a peer-reviewed paper and not an independently audited security evaluation.
Real, reported: based on the official article and the contemporaneous coverage cited below. Nothing added or dramatized. Sources checked November 14, 2026, at 7:15 AM ET (EST). Benchmarks are not the story here; the claim is that customers can now run more of the Managed Agents service on infrastructure they operate, in a private beta.
What Anthropic announced
The official article is titled "Operate Claude Managed Agents inside your security perimeter." It opens: "Introducing self-hosted Claude Managed Agents. Now you can run agents in your own environment, whether that’s in your own cloud or on-premise. You host the agents, credentials, and session history, and choose where requests to Claude are sent."
Anthropic said the option "is now available in private beta." The company described Managed Agents as "a suite of composable APIs for building and deploying autonomous agents" that already provided "the agent harness, sandboxes, credential handling, and session history." It said Managed Agents first launched as a hosted service on the Claude Platform, "where Anthropic runs the infrastructure for you," and that the company later added self-hosted sandboxes "for teams who needed their code to stay in their own environment."
The official post said agents are now "taking on more critical work, like fixing broken builds, testing releases, triaging alerts, and working through your bug backlog." For agents with that level of access, "some teams want the whole service in their own environment, not just the sandbox. Until now, that meant building the platform themselves." The new option, it said, puts "every part of the service" in the customer's cloud or on-premise environment, including the harness, sandboxes, credentials, and session history.
"You operate the Managed Agents server, its database, and the workers where agents run their shell and file tools." Model requests, the official post said, "go where you choose, whether that's Anthropic directly, your cloud provider (Amazon Bedrock, Google's Agent Platform, or Microsoft Foundry), or your own gateway." "Your developers still build agents with the same core APIs as the hosted Managed Agents service." Session history "is stored in your own database." If an organization has a zero data retention agreement with Anthropic, the company said that agreement "applies to the requests sent to Claude in the same way as any other Messages API call."
The official "Get started" section repeats that the option "is now available in private beta" and tells readers they can request access through an Anthropic account team or read the product documentation. Help Net Security, writing from the announcement, likewise said customers request access through their Anthropic account team, and it identified the session-history database as PostgreSQL. That database engine is Help Net Security's account; the official article says only "its database." Anthropic's product documentation gives the same route: "To request access to self-hosted Claude Managed Agents, contact your Anthropic account team."
Help Net Security summarized the earlier split in the product this way: customers "could host the sandbox themselves. That is the isolated environment where an agent’s code runs. Anthropic ran everything else, including the credential handling and the record of what each agent did." The official post is the source for the new, fuller self-hosted layout; the Help Net Security wording is the outlet's explanation of what changed.
What still leaves the customer's network
The official post is explicit that model inference is still a request sent to a provider the customer selects. Help Net Security stated the same limit in its own words: "Claude itself still runs outside the customer’s network, so prompts go out and completions come back." Anthropic's product documentation puts it this way: "The model runs at Anthropic or at a provider that you choose." The Help Net Security line is a press restatement of the official design, not a test we ran.
Anthropic also described four security principles that, it said, apply whether it hosts the service, the customer hosts only the sandbox, or the customer hosts the whole service. In the company's words, "the limits have to be built into the system itself." It said the harness that calls Claude is kept separate from the sandbox where the agent's code runs; "the agent loop, session history, and vault secrets stay outside it." Separating those parts, the post said, also means "an agent doesn't depend on one sandbox surviving its whole run, sandboxes start only when a tool needs them, and session history lives outside the sandbox as an event log."
Credentials are kept out of the sandbox, with a proxy adding them for MCP tools and "supported API calls," and self-hosted customers "can also store credentials in your own secrets manager." Each environment can be limited to "a list of allowed hosts," and "requests to any other host are blocked." The official example is an agent that fixes broken builds, whose list "might be just your git host and your package registry." When the customer hosts the sandbox, the post said those same limits are set in the customer's own firewall. Every session leaves a record of tool calls and results that can be exported over OpenTelemetry. Help Net Security noted that the proxy language is limited to MCP tools and supported API calls, "so any other service an agent reaches is worth checking before the agent goes live."
How to read the claims
This is an Anthropic product announcement for a private beta. Availability, the list of supported inference destinations, and the security architecture are company claims. The Help Net Security article summarizes the official post; it does not report an independent audit of the self-hosted build, and we did not test the software. A self-hosted control plane is also not the same thing as running the Claude model itself inside a customer's network. The official article frames the change as where the harness, credentials, and session history live, not as a new published model or a new benchmark.
Our advice (general guidance)
If you are considering this option, read the official article and the current documentation, not a recap, before you change how agents reach source code, credentials, or production systems. Self-hosting shifts operational work onto the customer, including the server, the database, and firewall or allow-list rules. Help Net Security made that point directly: those parts "are now the customer’s to run, and so is the job of keeping them patched and set up correctly."
Treat model prompts and completions as data that may still travel to a third-party inference provider, even when session history and credentials stay on your infrastructure. Confirm what a zero data retention agreement actually covers for those requests, and review which tools an agent can call beyond the proxy-backed MCP and supported API paths Anthropic described. This announcement is not a reason to paste production secrets into an unofficial setup guide or to approve access because a message cites a "private beta."
These are editorial precautions, not individualized legal, financial or security advice.
Sources
- Operate Claude Managed Agents inside your security perimeter (Anthropic, Nov 12, 2026)
- Self-hosted Claude Managed Agents keeps agent credentials in your environment (Help Net Security, Anamarija Pogorelec, Nov 12, 2026)
- Self-hosted Claude Managed Agents (Anthropic product documentation)
Published November 14, 2026 by FraudMeNot Team. Label: Real, reported.