← Voltar ao feed

Anthropic launches self-hosted Claude Managed Agents for in-environment credential control and session management

1 fonte(s) · 1 atualização(ões) · publicado 12/11/2026 15:07 · coletado há 51 minutos

Anthropic has introduced a self-hosted private beta of its Claude Managed Agents, enabling customers to run the agent harness, sandboxes, credentials, and session history entirely within their own cloud or on-premises environments. Previously, only the sandbox was self-hosted while credential management and session records were Anthropic-hosted. This shift allows organizations greater control over their environment and data, including credential management through their own secrets manager and firewall-enforced host allowlists to limit external communications. Model inference calls to Claude remain external but can be routed via various providers or customer gateways, with zero data retention agreements available. Anthropic's design enforces strict separation between the harness, sandbox, and credential storage to maintain security, with detailed logging via OpenTelemetry for integration into existing monitoring systems. Customers are now responsible for operating and securing the server, database, and firewall configurations.

Linha do tempo

  1. helpnetsecurity.com ↗ · pub. 12/11/2026 15:07 · coletado 12/11 17:01 primeiro registro

    Anthropic launches self-hosted Claude Managed Agents for in-environment credential control and session management

    ler a análise desta fonte

    Anthropic has introduced a self-hosted private beta of its Claude Managed Agents, enabling customers to run the agent harness, sandboxes, credentials, and session history entirely within their own cloud or on-premises environments. Previously, only the sandbox was self-hosted while credential management and session records were Anthropic-hosted. This shift allows organizations greater control over their environment and data, including credential management through their own secrets manager and firewall-enforced host allowlists to limit external communications. Model inference calls to Claude remain external but can be routed via various providers or customer gateways, with zero data retention agreements available. Anthropic's design enforces strict separation between the harness, sandbox, and credential storage to maintain security, with detailed logging via OpenTelemetry for integration into existing monitoring systems. Customers are now responsible for operating and securing the server, database, and firewall configurations.