IDE AI Coding Assistant Security Standard
Purpose: Set controls for AI coding assistants and AI-enabled developer environments.
Audience: Developers, DevSecOps, application security, security engineering, AI product developers, audit.
Primary Risks
- Source code exposure.
- Credential and secret leakage.
- Infrastructure information disclosure.
- AI-generated insecure code.
- Unauthorized AI coding assistant installation.
Control Requirements
| Requirement | Control ID | Description | Evidence |
|---|---|---|---|
| IDE Discovery | AI-CTRL-004, AI-CTRL-001 | Identify approved and unapproved AI coding tools and extensions. | IDE inventory, endpoint reports |
| Source Code Protection | AI-CTRL-004 | Detect submission of proprietary source code to AI assistants. | Detection logs, test evidence |
| Secret Detection | AI-CTRL-004 | Detect API keys, tokens, credentials, certificates and connection strings. | Secret detection alerts |
| Prompt Inspection | AI-CTRL-002 | Inspect developer prompts where technically feasible. | Prompt logs or policy events |
| Policy Enforcement | AI-CTRL-010 | Monitor, warn, redact, block or investigate risky submissions. | Policy screenshots, alert logs |
| Auditability | AI-CTRL-008 | Attribute activity to user, device, IDE and AI assistant. | Exportable audit records |
| SIEM Integration | AI-CTRL-009 | Forward events to SOC monitoring. | SIEM event ID, parser mapping |
Developer AI Coding Rules
Developers must not submit:
- Production credentials.
- API keys, tokens or certificates.
- Complete proprietary source files.
- Internal application logic not approved for external processing.
- Infrastructure-as-code templates containing sensitive infrastructure details.
- Customer, employee or regulated data.
- Security configurations, firewall rules, detection logic or privileged workflow details.
All AI-generated code must be reviewed through secure coding, peer review and DevSecOps processes before use.
PoC Tests
- Submit proprietary source code to AI coding assistant.
- Attempt to share API keys or tokens.
- Upload infrastructure-as-code template.
- Install unapproved AI extension.
- Request generation of privileged system code.