# AI Coding Tools — Acceptable Use Policy
Version: 1.0 | Effective: [DATE] | Owner: [SECURITY TEAM]
 
## Purpose
This policy governs the use of AI-assisted coding tools (including but not limited to 
GitHub Copilot, Claude Code, and Cursor) by engineering staff.
 
## Permitted Uses
- Code completion and generation for company projects
- Code review, refactoring, and debugging assistance
- Documentation and test generation
- Learning and skills development
 
## Prohibited Uses
- Sending customer PII, PHI, or financial data to AI APIs
- Sharing proprietary algorithms classified as trade secrets
- Bypassing code review requirements by attributing AI output as human-reviewed
- Using personal AI tool accounts for company work (use company-provisioned accounts)
- Submitting AI-generated code to open source projects without disclosure per project norms
 
## Data Classification
| Data Type | May Send to AI? |
|---|---|
| Public documentation | Yes |
| Internal code (no trade secrets) | Yes, with approved tools |
| Customer PII / PHI | No |
| Financial data | No |
| Security credentials / keys | No |
| Trade-secret algorithms | No — escalate to security team |
 
## Code Review Requirements
AI-generated code is subject to the same review requirements as human-written code. 
Engineers are responsible for understanding and verifying all code they commit, 
regardless of how it was generated.
 
## Security Requirements
- All secrets must be in approved secrets managers, never in prompts
- Report suspected prompt injection incidents to security@[company].com
- Report AI-suggested code that appears malicious to security@[company].com
 
## Compliance
For regulated workloads (SOC 2, HIPAA, PCI-DSS, ISO 27001), consult the security team 
before using AI tools. Additional data residency and audit requirements may apply.
 
## Violations
Violations of this policy may result in access revocation and disciplinary action 
per the standard HR process.

# AI Coding Tools Security Baseline
Complete before granting team access.
 
## Pre-Deployment
- [ ] Vendor DPA (Data Processing Agreement) reviewed and signed
- [ ] Data residency requirements confirmed (EU? US only?)
- [ ] Commercial plan terms confirmed (Claude Team/Enterprise and API: no training on your data by default; verify current terms with the vendor)
- [ ] API key management via secrets manager (not .env files in repos)
- [ ] Network policy updated if outbound API calls need allowlisting
 
## IDE / Tool Configuration
- [ ] Copilot: confirm the data-sharing / product-improvement setting is disabled at the org level
  (exact menu names change — check current GitHub docs)
- [ ] Claude Code: deploy a managed-settings.json (see Template 5) so permission, sandbox, and telemetry settings are enforced rather than left to each developer
- [ ] Cursor: Disable codebase indexing / enable Privacy Mode if code sensitivity requires it
  (exact menu names change — check current Cursor docs)
 
## Repository Hardening  
- [ ] .gitignore includes: .env, *.key, *.pem, secrets.*, credentials.*
- [ ] git-secrets or detect-secrets pre-commit hook installed
- [ ] Branch protection rules prevent direct push to main
- [ ] CLAUDE.md / copilot-instructions committed with no-secrets guidance (guidance only — enforce with deny rules and the sandbox, see Template 5)
 
## Ongoing
- [ ] Monthly review of API usage logs for anomalies
- [ ] Quarterly policy review for new tool features that may change data exposure
- [ ] Annual penetration test includes AI tool integrations

# AI Tool Procurement Evaluation Checklist
 
## Data Privacy
- [ ] Does the vendor train on customer code? (Answer must be: No for enterprise tier)
- [ ] Where is data processed? (Region / data residency)
- [ ] Is there a Data Processing Agreement available?
- [ ] What is the data retention period for prompts and completions?
- [ ] Can we audit what data was sent to the vendor?
 
## Security
- [ ] SOC 2 Type II report available?
- [ ] Pen test results available under NDA?
- [ ] SSO / SAML support for identity management?
- [ ] Role-based access control (admin vs. user)?
- [ ] Audit logs available for admin review?
 
## Compliance
- [ ] GDPR compliant (if EU data involved)?
- [ ] HIPAA BAA available (if healthcare data)?
- [ ] FedRAMP authorized (if US government work)?
- [ ] ISO 27001 certified?
 
## Operability
- [ ] API available for CI/CD integration?
- [ ] Deployment options (Claude Code can run through Amazon Bedrock, Google Vertex AI, or Microsoft Foundry on your own cloud account)?
- [ ] SLA uptime guarantee?
- [ ] Support tier during business hours?
 
## Commercial
- [ ] Per-seat vs. usage-based pricing?
- [ ] Volume discounts available?
- [ ] Contract term flexibility (monthly vs. annual)?
- [ ] Exit / data portability terms?
 
## Vendor Health
- [ ] Funding / runway transparency?
- [ ] Support channel response time?
- [ ] Roadmap published and aligned with our needs?

# AI Tool Security Incident Response
 
## Trigger Conditions
Invoke this procedure if:
1. AI tool suggests code containing a hardcoded secret
2. AI output appears to include another organization's proprietary code
3. Suspected prompt injection in code comments or AI outputs
4. AI tool account credentials compromised
 
## Response Steps
 
### Immediate (within 1 hour)
1. Revoke the affected API key / access token immediately
2. Scan recently committed code for AI-suggested content that may be affected
3. Notify security@[company].com
 
### Investigation (within 24 hours)
1. Identify all repositories where affected tool was used in last 30 days
2. Run secret scanning across those repos: including history (for example `gitleaks detect` or `trufflehog git file://.`)
3. Check vendor's audit logs if available
 
### Remediation
1. Rotate any exposed credentials (assume compromised)
2. If secrets were committed, rewrite history only after rotation (git-filter-repo or BFG) and coordinate any force-push with the team. Rotation, not history rewriting, is what contains the exposure.
 
### Post-Incident
1. Root cause analysis: how did AI suggest this content?
2. Update CLAUDE.md / copilot-instructions to prevent recurrence
3. Notify affected parties if customer data was involved (legal/compliance review)
4. Publish internal post-mortem within 5 business days

{
  "permissions": {
    "disableBypassPermissionsMode": "disable",
    "disableAutoMode": "disable",
    "deny": [
      "Read(./.env)",
      "Read(./.env.*)",
      "Read(./secrets/**)",
      "Bash(curl *)"
    ]
  },
  "allowManagedPermissionRulesOnly": true,
  "allowManagedHooksOnly": true,
  "allowManagedMcpServersOnly": true,
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_METRICS_EXPORTER": "otlp",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "grpc",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "https://otel-collector.example.com:4317"
  }
}