Codex: Codex Security & Enterprise Management
Last updated: 2026-08-31
Using Codex in enterprise environments requires consideration of security, compliance, and management. This lesson covers enterprise-grade security strategies and best practices.
📋 Prerequisites: Understanding Codex basic configuration and permission settings
1. What You Will Learn
- Data privacy and security
- Auditing and compliance
- Enterprise features
- Management strategies
2. Data Privacy
(1) Data Flow
| Scenario | Local Data | Cloud Data |
|---|---|---|
| CLI | Code stays local | Prompt + context sent to API |
| Web/App | Code cloned to sandbox | Code processed in cloud sandbox |
| IDE | Code stays local | Selected code + context sent to API |
(2) OpenAI Data Policy
| Policy | Description |
|---|---|
| No Training | API data not used for model training |
| Retention Period | Data retained 30 days for abuse review |
| Encryption | Encrypted in transit and at rest |
| Access Control | Strictly limited internal access |
(3) Sensitive Data Handling Recommendations
| Data Type | Recommendation |
|---|---|
| API Keys | Add to blocked_paths, prevent Codex from reading |
| User Data | Use anonymized datasets |
| Trade Secrets | Use Business/Enterprise edition with EKM |
| Personal Information | Comply with GDPR/data protection laws |
3. Enterprise Features
(1) Plan Comparison
| Feature | Business | Enterprise |
|---|---|---|
| SAML SSO | ✅ | ✅ |
| MFA | ✅ | ✅ |
| SCIM User Management | ❌ | ✅ |
| EKM (Encryption Key Management) | ❌ | ✅ |
| RBAC | ❌ | ✅ |
| Audit Logs | Basic | Complete |
| Data Residency | ❌ | ✅ |
| Larger VMs | ✅ | ✅ |
| Custom Retention | ❌ | ✅ |
(2) SSO Integration
TEXT
📖 Display only
Enterprise edition supports SAML SSO with existing identity systems:
1. Configure SAML IdP (Okta, Azure AD, OneLogin, etc.)
2. Set up SSO in the OpenAI enterprise dashboard
3. Employees log in to Codex with corporate accounts
4. Unified identity management
(3) EKM (Encryption Key Management)
Enterprise edition supports EKM — the enterprise manages its own encryption keys:
TEXT
📖 Display only
1. OpenAI cannot decrypt enterprise data
2. Keys managed by enterprise KMS
3. Even if OpenAI is breached, data cannot be read
4. Suitable for highly sensitive scenarios
4. Auditing & Compliance
(1) Audit Logs
TOML
[audit]
log_all_operations = true
log_file = "/var/log/codex/audit.log"
include_shell_output = true
include_file_diffs = true
retention_days = 90
Audit log contents:
TEXT
📖 Display only
[2026-08-28 10:30:15] USER=alice ACTION=read FILE=src/auth.ts
[2026-08-28 10:30:18] USER=alice ACTION=write FILE=src/auth.ts DIFF=+5,-2
[2026-08-28 10:30:22] USER=alice ACTION=shell CMD="npm test" EXIT=0
[2026-08-28 10:30:25] USER=alice ACTION=network URL=https://registry.npmjs.org
(2) Compliance Requirements
| Standard | Description | Codex Support |
|---|---|---|
| SOC 2 | Security, availability, confidentiality | Enterprise edition |
| GDPR | EU data protection | Data residency |
| HIPAA | Medical data protection | BAA agreement |
| ISO 27001 | Information security management | Enterprise edition |
5. Management Strategies
(1) Roles & Permissions
| Role | Permissions | Best For |
|---|---|---|
| Admin | Full permissions + user management | IT administrators |
| Developer | Project read/write + sandbox operations | Development teams |
| Reviewer | Read-only + code review | Code review teams |
| Guest | Read-only | External collaborators |
(2) Usage Policies
TOML
# Enterprise policy configuration
[policy]
# Blocked models
blocked_models = ["gpt-3.5-turbo"]
# Enforced sandbox mode
enforce_sandbox = "workspace-write"
# Blocked operations
blocked_operations = [
"network_access",
"system_path_access",
]
# Max parallel tasks
max_parallel_tasks = 5
# Audit requirements
audit_all = true
▶ Example 1: Alice's Enterprise Configuration
TOML
# Alice's company enterprise policy
[policy]
enforce_sandbox = "workspace-write"
blocked_paths = [".env", "secrets/", "credentials/"]
blocked_operations = ["network_access"]
audit_all = true
retention_days = 180
[approval]
shell_commands = "ask"
file_writes = "ask"
network_access = "deny"
6. Security Best Practices
| Practice | Description |
|---|---|
| Least Privilege | Only authorize necessary permissions |
| Audit Everything | Enable complete audit logs |
| Regular Reviews | Check Codex operation records |
| Key Isolation | Add API Keys and key files to protection lists |
| Network Restrictions | Deny network access in production environments |
| Data Residency | Require local storage for sensitive data |
| Employee Training | Train proper Codex usage |
❓ FAQ
Q How much more does Enterprise cost?
A Business is pay-as-you-go, Enterprise requires contacting sales. Specific pricing depends on user count and usage.
Q How does EKM work?
A The enterprise uses its own KMS to manage encryption keys. Data stored by OpenAI is encrypted with enterprise keys, so OpenAI cannot decrypt it.
Q Can Codex operations be undone?
A Code modifications can be rolled back via git. API calls cannot be undone; sent data cannot be recalled.
Q How do I ensure code isn't used for training?
A Use Business/Enterprise edition — API data is not used for training. Free/Plus API also doesn't train, but data is retained for 30 days.
Q Can I restrict Codex to only use specific models?
A Yes. Set
blocked_models or enforce_model in policy configuration.📖 Summary
- Data privacy: CLI processes locally, Web/App uses cloud sandbox
- Enterprise edition: SSO, EKM, auditing, data residency
- Audit logs: record all operations, support compliance
- Management strategies: role permissions, operation restrictions, enforced sandbox
- Best practices: least privilege + auditing + key isolation
📝 Exercises
- Basic (⭐): Configure audit logs to record all Codex operations.
- Intermediate (⭐⭐): Design an enterprise-grade security policy configuration with permission control and sensitive data protection.
- Advanced (⭐⭐⭐): Write a Codex enterprise usage security guide covering data privacy, compliance, auditing, and incident response.