---
title: "Claude Code enterprise security checklist | Aventail™"
url: https://aventail.co.uk/guides/claude-code-security-enterprise
description: "Review Claude Code permissions, inference routing, fallback rules and telemetry before rolling out private AI coding across your enterprise."
lang: en-GB
---

Skip to content

Locai is now Aventail™ Learn about the rebrand: https://aventail.co.uk/newsroom/locai-rebrands-as-aventail

All guides (https://aventail.co.uk/guides)

# Claude Code security for enterprises: a practical deployment checklist

26 October 2026

Review Claude Code permissions, inference routing, fallback rules and telemetry before rolling out private AI coding across your enterprise.

You can keep Claude Code’s model requests on approved infrastructure. That does not, by itself, control the files the agent reads, the commands it runs or every connection the application makes. Review those boundaries separately before introducing confidential code.

This guide is for the engineering lead planning a rollout and the IT or security team approving it. The aim is a small, testable deployment with evidence of where requests ran, rather than a promise that one privacy setting solves everything.

## Separate the three boundaries

### What the coding agent can access

A coding agent can read project files and return tool results to its model. Give it a clean working copy with the files needed for the task, without production credentials, customer exports or unrelated repositories. Consider what build scripts, plugins and MCP servers can access too.

Claude Code’s permission controls (https://code.claude.com/docs/en/permissions) govern actions such as editing files and executing commands. Review the effective configuration for the installed version and session mode. A model routing rule is not a replacement for those controls.

### Where model requests are processed

Requests may contain the user’s prompt, selected source files, tool output and conversation context. Decide which of those may be processed on the developer’s device, an internal server or an approved external provider. Apply the same restriction to fallback routes.

“The model is local” is an incomplete description if a fallback can send the next request elsewhere. Write the requirement as an allowed destination list and a failure condition: for example, “Only this device may process the request; if it cannot, stop.”

### What the application and its tools send or store

Review session history, diagnostic uploads, updates, authentication and optional integrations independently of inference. Anthropic’s data usage documentation (https://code.claude.com/docs/en/data-usage) distinguishes model traffic, local session storage and other data flows. Check the terms and settings that apply to your deployment.

Changing the model endpoint does not prove that the entire application has become offline or that no data reaches another service. If your requirement is zero external traffic, your network and application review must establish that separately.

## Define an approval you can actually test

Before choosing a model, agree these points with the person responsible for the code:

- **Allowed data:** a named repository or class of information, with excluded files and secrets identified.
- **Allowed processing:** this device only, named internal machines, or specified external provider accounts. State any geographical restrictions.
- **Allowed actions:** whether the agent may edit files, install dependencies, run shell commands or use external tools.
- **Failure behaviour:** which alternative destinations are permitted, and when the request must fail.
- **Evidence:** application and runtime versions, model identifier, effective policy, request destination and the person approving the pilot.

For example, a first pilot could allow an assistant to explain and test a synthetic repository on a developer’s laptop, with no remote inference route. That is easier to verify than approving every company repository and provider at once.

## Configure a local-only Aventail ™ deployment

For the supported Connect path, install Aventail ™ Link, serve a compatible model and enable Claude Code in Connected Apps. Choose the model in Aventail ™ Link, then open a fresh Claude Code session. Follow the Claude setup guide (https://aventail.co.uk/guides/claude-code-with-aventail-connect) for the current interface and prerequisites.

To keep model processing on that device, permit only **On device** in the applicable routing policy. Check the effective rules, including inherited policies and fallback paths: a peer or provider route would expand the permitted boundary. If no destination is permitted or available, the request fails rather than going to a disallowed destination. The routing and verification guide (https://aventail.co.uk/guides/route-local-models-and-cloud-apis) explains how to check the actual execution destination in Recent usage.

If the policy allows an internal server instead, name that machine and validate its access controls before using real code. Aventail ™ ’s security documentation (https://docs.aventail.co.uk/docs/guides/security) makes an important distinction: signing in to Aventail ™ Control does not automatically protect a model’s serving port. Start with loopback access for a same-machine pilot; have your infrastructure owner review authentication, transport and network rules before allowing other machines to connect.

Aventail ™ Control manages configuration outside the inference request path. Check architecture and data flow (https://docs.aventail.co.uk/docs/introduction/platform-architecture) and usage analytics and privacy (https://docs.aventail.co.uk/docs/privacy/usage-analytics) when reviewing operational reporting. The product-analytics switch is not a switch for all device reporting or application telemetry.

## Keep action permissions separate from routing

Even a model running on an approved device can propose a destructive command or follow an instruction embedded in a file. Inspect untrusted repositories and tool output as inputs, not as authority to change your security policy.

Start with permissions that require review for consequential edits and commands. Avoid using permission-bypass options as the normal enterprise setup. Review broad allow rules and who can change them.

Where appropriate, add an operating-system boundary such as a restricted development container. Claude Code’s sandbox documentation (https://code.claude.com/docs/en/sandboxing) explains the scope of its shell sandbox and the components outside it. Confirm the restrictions work on your chosen operating system; do not assume a shell sandbox confines every plugin or external service.

## Run this acceptance test before using confidential code

Use a small synthetic repository. Do not test with real secrets or customer data.

1. **Prove the intended path.** Ask the assistant to explain a harmless function. Check the request’s actual model, machine and routing outcome in Recent usage; a model name displayed in the app is not sufficient evidence.
2. **Prove the stop condition.** With only local execution permitted, stop serving the local model and submit another harmless request. It should fail. An unexpected remote success means the policy or configuration needs investigation.
3. **Check the wider network boundary.** Review application and endpoint network records for the pilot. Account for connections that do not represent inference, rather than assuming all traffic uses the runtime.
4. **Check an action boundary.** Use a harmless edit in a disposable working copy to verify the expected approval behaviour. Confirm the person reviewing the change can see what will happen before approving it.
5. **Check storage and reporting.** Inspect where sessions and diagnostic records are retained, who can access them and how they are removed. Avoid copying full prompts or code into a broadly accessible review document.
6. **Keep a short evidence record.** Record versions, the synthetic test, the effective rule, the observed destination and any failure. Repeat the relevant checks after changing the app, model, runtime or policy.

An inference record helps establish the model request’s destination. It does not, on its own, establish everything the application or its tools transmitted.

## Decide whether to widen the pilot

Expand only after the permitted task works and the failure tests behave as intended. For a team rollout, assign an owner for policy distribution, application configuration, model updates and removing departing users or devices. The private workspace deployment guide (https://aventail.co.uk/guides/private-ai-workspace-deployment) covers a staged rollout across laptops and servers.

If an external model is permitted for some work, document the allowed data, provider account and fallback conditions. Do not treat “not used for training” and “not sent to an external provider” as the same requirement.

## Practical questions

### Can Claude Code use confidential source code without sending model requests to a public API?

Yes, for a supported application and model configuration that keeps processing on approved compute. Configure and test the effective routing policy, then review application storage, permissions and other network activity separately. This does not mean Anthropic’s Claude model weights run locally.

### Does Aventail ™ Connect control every connection Claude Code makes?

No. Aventail ™ Connect brings supported model requests into Aventail ™ Link. File access, shell actions, telemetry and other app or tool connections require their own controls.

### Is a successful local response enough for security approval?

No. Check what happens when local serving is unavailable, inspect inherited and fallback rules, and review the other two boundaries. Use Security & trust (https://aventail.co.uk/security) for Aventail ™ ’s security evidence alongside your deployment assessment.

Technical references checked on 26 October 2026. Application behaviour and controls can change; retain the versions used for your assessment.

We use optional cookies and tracking to identify visiting companies and load our contact form. Cookie policy (https://aventail.co.uk/cookies)

## Structured data

```json
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://aventail.co.uk/#organization",
      "name": "Aventail",
      "alternateName": "Locai",
      "legalName": "Loc.ai Ltd",
      "description": "Aventail, formerly Locai, was founded in May 2025 and is a graduate of Google AI First. Its integrated AI inference platform combines a built-in inference engine, deployment of supported models across approved hardware, and policy-based routing between those machines and approved external AI providers. Organisations use Aventail to control where inference requests run.",
      "url": "https://aventail.co.uk/",
      "logo": "https://aventail.co.uk/apple-touch-icon.png",
      "email": "hello@aventail.co.uk"
    },
    {
      "@type": "Article",
      "@id": "https://aventail.co.uk/guides/claude-code-security-enterprise#page",
      "url": "https://aventail.co.uk/guides/claude-code-security-enterprise",
      "name": "Claude Code security for enterprises: a practical deployment checklist",
      "description": "Review Claude Code permissions, inference routing, fallback rules and telemetry before rolling out private AI coding across your enterprise.",
      "publisher": {
        "@id": "https://aventail.co.uk/#organization"
      },
      "headline": "Claude Code security for enterprises: a practical deployment checklist",
      "datePublished": "2026-10-26",
      "dateModified": "2026-10-27T07:29:58.101634+00:00",
      "image": "https://aventail-ai-platform.lovable.app/images/share/newsroom/claude-code-security-enterprise-v1.jpg",
      "author": {
        "@id": "https://aventail.co.uk/#organization"
      }
    },
    {
      "@type": "BreadcrumbList",
      "@id": "https://aventail.co.uk/guides/claude-code-security-enterprise#breadcrumb",
      "itemListElement": [
        {
          "@type": "ListItem",
          "position": 1,
          "name": "Home",
          "item": "https://aventail.co.uk/"
        },
        {
          "@type": "ListItem",
          "position": 2,
          "name": "Guides",
          "item": "https://aventail.co.uk/guides"
        },
        {
          "@type": "ListItem",
          "position": 3,
          "name": "Claude Code security for enterprises: a practical deployment checklist",
          "item": "https://aventail.co.uk/guides/claude-code-security-enterprise"
        }
      ]
    }
  ]
}
```