Cisco DefenseClaw pairs with NVIDIA OpenShell sandbox
Cisco described DefenseClaw on 21 Oct 2026 as a security layer that checks every prompt, tool call and tool result for a coding agent in an NVIDIA OpenShell sandbox, with directory groups choosing the policy profile. Cisco says it decides each web destination by host name and exports decisions to Splunk or OpenTelemetry.
Facts read on .
On 21 Oct 2026, Cisco published a blog post on protecting enterprise coding agents with DefenseClaw and NVIDIA OpenShell. Cisco says DefenseClaw connects OpenShell's sandbox controls with identity-aware policy, AI Defense guardrails and agent observability. The user's directory group picks a policy profile, the profile shapes the sandbox and its egress policy, and guardrail hooks are relayed to the DefenseClaw gateway.
Cisco says DefenseClaw checks every prompt, tool call and tool result using its own rules and Cisco AI Defense. It decides each web destination for a sandboxed agent by host name without decrypting traffic, and the sandbox supplies provider credentials only for authorized requests so the agent sees placeholders. Administrators can set user- or group-based policy for which agents run on managed endpoints and which run only in a sandbox.
Cisco states that every block, approval and decision is recorded with the agent, sandbox and policy behind it and can be sent to Splunk or any OpenTelemetry backend. These are Cisco's descriptions.
Sources
- Cisco: Protect enterprise coding agents with Cisco DefenseClaw and NVIDIA OpenShell (blogs.cisco.com), read