Sketchnote diagram for: Codex CLI Enterprise Admin Setup: RBAC, Managed Configuration, and Compliance APIs

Codex CLI Enterprise Admin Setup: RBAC, Managed Configuration, and Compliance APIs

On 14 May 2026, OpenAI published the first dedicated Enterprise Admin Setup guide for Codex, consolidating workspace enablement, RBAC, managed configuration, access tokens, and governance APIs into a single reference ¹. This article walks through the seven-step rollout process, explains how managed configuration layers interact with local settings, and shows how the Compliance and Analytics APIs close the observability loop for regulated environments.

Why a Dedicated Admin Guide Now?

Codex has shipped hooks general availability, access tokens for non-interactive workflows, and device-code authentication in the same release window ². Each feature creates new surface area that enterprise security teams need to govern. Before May 2026, administrators assembled rollout procedures from scattered docs on sandbox modes, approval policies, and workspace settings. The new guide unifies these into a prescriptive sequence.

The Three Owners Model

OpenAI recommends identifying three organisational roles before touching any settings ¹:

Role	Responsibility
ChatGPT Enterprise Workspace Owner	Enables Codex surfaces, configures workspace toggles
Security Owner	Sets agent permission boundaries, approval policies, sandbox constraints
Analytics Owner	Integrates Analytics and Compliance APIs, monitors governance data

This separation of duties mirrors traditional cloud landing-zone patterns where infrastructure, security, and observability each have distinct owners.

Step-by-Step Rollout

1. Enable Codex Surfaces

Codex Local (the CLI and desktop app) is enabled by default on new ChatGPT Enterprise workspaces ¹. Codex Cloud requires explicit enablement and GitHub repository connection. Toggle both from Workspace Settings → Settings and Permissions.

Two optional toggles matter for CLI-centric teams:

Access tokens — allow members to run Codex from scripts, schedulers, and private CI runners using their ChatGPT workspace identity ²
Device-code authentication — supports non-interactive CLI environments such as SSH sessions and containers where browser-based OAuth is impractical ¹

2. Configure RBAC

Codex RBAC sits inside ChatGPT Enterprise’s existing custom-role system. Permissions resolve to the most permissive (least restrictive) access across all assigned roles ¹.

OpenAI recommends two groups:

Codex Users    → general access, inherits default policy
Codex Admin    → settings management, policy deployment, analytics

The Codex Admin role grants access to the managed-configuration page, analytics dashboards, cloud-managed policies, and GitHub connector management ¹. If your organisation uses SCIM, back the Codex Admin group with your identity provider so membership changes are auditable and centrally managed ¹.

graph TD
    IDP[Identity Provider<br/>SSO / SCIM] --> WS[ChatGPT Enterprise Workspace]
    WS --> CU[Codex Users Group<br/>Default policy]
    WS --> CA[Codex Admin Group<br/>Admin policy]
    CA --> MC[Managed Configuration]
    CA --> AN[Analytics API]
    CA --> CO[Compliance API]
    CU --> CLI[Codex CLI]
    CU --> APP[Codex Desktop App]
    CU --> IDE[IDE Extension]

3. Deploy Managed Configuration

Managed configuration is the mechanism that makes enterprise policy enforceable rather than advisory. It has two layers ³:

Requirements (requirements.toml) — admin-enforced constraints that users cannot override
Managed defaults (managed_config.toml) — starting values applied at launch; users may modify them during sessions

Requirements Precedence

Codex applies requirements in a first-match-wins sequence ³:

Cloud-managed requirements (ChatGPT Business/Enterprise dashboard)
macOS MDM via com.openai.codex:requirements_toml_base64
System requirements.toml (/etc/codex/ on Unix; %ProgramData%\OpenAI\Codex\ on Windows)

Example: Locking Down a Regulated Environment

# /etc/codex/requirements.toml
allowed_approval_policies = ["untrusted", "on-request"]
allowed_sandbox_modes = ["read-only", "workspace-write"]
allowed_web_search_modes = ["cached"]

[features]
browser_use = false
computer_use = false

[rules]
prefix_rules = [
  { pattern = [{ token = "rm" }], decision = "forbidden" },
  { pattern = [{ token = "git", subtokens = ["push", "force"] }], decision = "forbidden" },
  { pattern = [{ token = "git" }], decision = "prompt" }
]

[experimental_network]
enabled = true
allowed_domains = ["api.openai.com", "registry.npmjs.org", "pypi.org"]
denied_domains = ["*"]

This configuration restricts approval policy to human-reviewed modes, disables browser and computer use, blocks destructive shell commands, and permits network access only to package registries ³.

Host-Specific Sandbox Rules

For teams with heterogeneous infrastructure — local laptops and remote devboxes — use [[remote_sandbox_config]] to apply different policies per host ³:

allowed_sandbox_modes = ["read-only"]

[[remote_sandbox_config]]
hostname_patterns = ["*.devbox.example.com"]
allowed_sandbox_modes = ["read-only", "workspace-write"]

MCP Server Allowlisting

Requirements can also restrict which MCP servers the CLI is permitted to connect to ³:

[mcp_servers.docs]
identity = { command = "codex-mcp" }

Any MCP server not listed in the allowlist is rejected at startup, preventing developers from introducing unvetted tool servers.

4. macOS MDM Deployment

For organisations using Jamf, Kandji, or Mosyle, managed configuration can be pushed as a device profile ³:

Build the TOML payload
Encode with base64
Add to MDM profile under com.openai.codex with key requirements_toml_base64 (for requirements) or config_toml_base64 (for defaults)
Push profile; users restart Codex
Verify startup configuration reflects managed values

This is the only deployment path that does not require users to accept or install anything — the profile lands silently via MDM.

5. Team Configuration via `.codex`

Below the enterprise layer sits repository-scoped team configuration ⁴. The .codex directory checked into a repository provides:

.codex/
  config.toml       # defaults: sandbox, approvals, model
  rules/            # command governance rules
  skills/           # shared agent capabilities

Team config applies to anyone who opens the repository, regardless of their personal ~/.codex/config.toml settings. Enterprise requirements always take precedence over team config ³.

graph TD
    subgraph Precedence["Configuration Precedence (highest → lowest)"]
        REQ[Enterprise Requirements<br/>Cloud / MDM / System]
        MD[Managed Defaults<br/>managed_config.toml]
        TC[Team Config<br/>.codex/config.toml]
        UC[User Config<br/>~/.codex/config.toml]
        FL[CLI Flags]
    end
    REQ --> MD --> TC --> UC --> FL
    style REQ fill:#d4edda
    style MD fill:#d4edda
    style TC fill:#fff3cd
    style UC fill:#f8d7da
    style FL fill:#f8d7da

⚠️ CLI flags sit at the bottom of the precedence stack — they are overridden by managed layers. This is a deliberate design choice ensuring that --full-auto on the command line cannot bypass an enterprise policy restricting approval to on-request.

6. Governance: Analytics and Compliance APIs

Two API surfaces close the observability loop ¹:

Analytics API

Scoped to codex.enterprise.analytics.read, the Analytics API exposes:

/workspaces/{workspace_id}/usage — token consumption, session counts, model distribution
/code_reviews — auto-review trigger counts, pass/fail rates
/code_review_responses — developer responses to review findings

Setup requires a dedicated read-only API key. Email support@openai.com with the key’s last four digits to request scoping ¹.

Compliance API

Scoped to read and delete permissions, the Compliance API provides:

/compliance/workspaces/{workspace_id}/logs — audit trail of Codex interactions
/codex_tasks — cloud task metadata
/codex_environments — environment configuration snapshots

These endpoints support SIEM integration, retention-policy enforcement, and right-to-erasure requests under GDPR ¹.

7. Verification Checklist

OpenAI’s guide concludes with a verification matrix ¹:

Local and cloud sign-in functions correctly
MFA and SSO requirements match corporate security policy
RBAC and workspace toggles produce expected access behaviour
Managed configuration applies (check codex doctor output for policy layer)
Governance data is visible to analytics owners

The codex doctor command, shipped in v0.131.0, now reports which managed-configuration layer is active, making it the fastest way to verify that enterprise policy has landed on a developer’s machine ⁵.

Security Guarantees

OpenAI documents the following enterprise security properties for Codex ⁶:

Property	Detail
Data training	Enterprise data is never used for model training
Data retention	Zero data retention for App, CLI, and IDE Extension
Encryption at rest	AES-256
Encryption in transit	TLS 1.2+
Audit logging	Via Compliance API
Residency and retention	Follows ChatGPT Enterprise workspace policies

Anti-Patterns

Granting Codex Admin to all developers. The admin role provides write access to managed policies that affect every user in the workspace. Keep it to platform engineering, IT, and governance operators.

Relying solely on team config for security. Repository-scoped .codex/config.toml is advisory — developers can override it locally. Use enterprise requirements for anything security-critical.

Skipping the Analytics API. Without usage telemetry, you cannot detect token-budget anomalies, shadow model usage, or unapproved MCP server connections.

Deploying full-auto via managed defaults. Even as a managed default, full-auto removes the human review checkpoint. For regulated environments, enforce on-request or untrusted via requirements instead.

Practical Rollout Sequence

For teams moving from ad-hoc Codex adoption to governed deployment:

sequenceDiagram
    participant PlatEng as Platform Engineering
    participant SecOwner as Security Owner
    participant DevTeam as Development Team
    participant API as Analytics/Compliance

    PlatEng->>PlatEng: Enable Codex surfaces in workspace
    PlatEng->>PlatEng: Create RBAC groups (Users + Admin)
    SecOwner->>SecOwner: Draft requirements.toml
    SecOwner->>PlatEng: Deploy via cloud dashboard or MDM
    PlatEng->>DevTeam: Announce rollout, distribute .codex team config
    DevTeam->>DevTeam: Run codex doctor to verify policy
    PlatEng->>API: Configure Analytics API key
    SecOwner->>API: Configure Compliance API key
    API->>PlatEng: Usage and audit data flows to SIEM

Conclusion

The enterprise admin setup guide brings Codex’s governance story to parity with what security teams expect from any developer tooling rollout: identity-backed access control, centrally enforced policy, and API-driven observability. The combination of cloud-managed requirements, MDM deployment, and the Compliance API makes it possible to adopt Codex CLI at scale without sacrificing the controls that regulated industries demand.

Citations

OpenAI, “Admin Setup – Codex”, OpenAI Developers, May 2026. https://developers.openai.com/codex/enterprise/admin-setup ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹
OpenAI, “Changelog – Codex”, OpenAI Developers, May 2026. https://developers.openai.com/codex/changelog ↩ ↩²
OpenAI, “Managed configuration – Codex”, OpenAI Developers, May 2026. https://developers.openai.com/codex/enterprise/managed-configuration ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷
OpenAI, “Config basics – Codex”, OpenAI Developers, 2026. https://developers.openai.com/codex/config-basic ↩
OpenAI, “Codex CLI v0.131.0 Release”, GitHub Releases, May 2026. https://github.com/openai/codex/releases ↩
OpenAI, “Security – Codex”, OpenAI Developers, 2026. https://developers.openai.com/codex/security ↩

Codex CLI Enterprise Admin Setup: RBAC, Managed Configuration, and Compliance APIs

Why a Dedicated Admin Guide Now?

The Three Owners Model

Step-by-Step Rollout

1. Enable Codex Surfaces

2. Configure RBAC

3. Deploy Managed Configuration

Requirements Precedence

Example: Locking Down a Regulated Environment

Host-Specific Sandbox Rules

MCP Server Allowlisting

4. macOS MDM Deployment

5. Team Configuration via .codex

6. Governance: Analytics and Compliance APIs

Analytics API

Compliance API

7. Verification Checklist

Security Guarantees

Anti-Patterns

Practical Rollout Sequence

Conclusion

Citations

5. Team Configuration via `.codex`