Discussion

Anthropic puts Claude Managed Agents inside customers’ own security perimeters

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#12598

Anthropic has opened a private beta for running Claude Managed Agents in a customer’s own cloud or on-premises environment. The key change is where the agent infrastructure and its records can live: within the organisation’s own systems, rather than solely on Anthropic’s infrastructure.

Anthropic Watch analysis

What happened

Anthropic says the self-hosted service includes the agent harness, sandboxes, credential management and session-history storage in the customer’s own database. The company describes it as an option for teams that need agents to work with internal systems while keeping infrastructure within their security perimeter.

The offer is currently a private beta, so this is not yet a generally available deployment option. Anthropic’s announcement describes its intended security and privacy benefits; those are product claims, not proof that every deployment will be secure by default.

Why it matters

For organisations considering unattended agents, the location of credentials, session records and execution environments is not a footnote. Keeping those components on premises or in a company’s own cloud could make it easier to fit agents into existing access controls and data policies. It also gives technical teams more infrastructure to operate and secure themselves.

That makes this a meaningful change in the deployment choices Anthropic is offering, not simply another agent feature. The practical test will be whether customers get useful control without inheriting a fresh pile of operational chores.

Our read

Self-hosting addresses a real sticking point for companies wary of sending agent infrastructure and session history outside their environment. But “inside the perimeter” is a deployment choice, not a magic security cloak. Teams weighing the beta should look closely at how the components are installed, updated and monitored before trusting them with sensitive systems.

What to watch

  • When the private beta expands or becomes generally available.
  • What deployment, maintenance and monitoring responsibilities fall to customers.
  • How organisations can control agent access to internal systems and credentials.

Discussion spark: Would keeping an agent’s infrastructure and session history in your own environment make you comfortable letting it work with internal systems, or would operational responsibility outweigh that benefit?

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.