SOP-010: AI Tool Usage & Data Handling

Version: 1.0 · Owner: AI Platform Owner + Security · Applies to: All engineers

Purpose & scope

Safe, compliant use of AI coding and planning tools (Cursor, Amazon Q, Bedrock agents, MCP).

Roles

Role	RACI
AIPO	A — platform config, audit
SEC	A — data classification rules
DEV	R — compliant daily use

Approved tools (primary stack)

Tool	Use	Data boundary
Cursor (enterprise)	IDE + agents	No prod PII; enterprise privacy mode
Amazon Q Developer	AWS API / IaC help	AWS enterprise boundary
Bedrock (via gateway)	Planning agents, batch tasks	VPC endpoint; logged prompts
MCP (internal)	ADR, spec, runbook search	Read-only; Accepted/Approved only

Optional (team discretion): local Ollama for offline snippets — no company code.

Procedure

Before use

1. Confirm enterprise AI agreements active; privacy mode on in Cursor.
2. Complete onboarding module (SOP-011 step 5).

During development

3. Context hygiene — Load Approved spec + Accepted ADRs via MCP; avoid pasting large prod log dumps.
4. No prohibited data in prompts — Prod PII, credentials, full customer records, unredacted security findings.
5. Synthetic fixtures only — Test data generated or from approved fixture library.
6. Verify AI output — Never merge unreviewed AI-generated code; treat as untrusted input.

Prohibited

• Pasting production database exports into AI chat
• Sending secrets, API keys, or .env contents to models
• Using personal ChatGPT accounts for company code
• Auto-accepting AI architectural decisions (→ SOP-002)

Audit

7. AIPO samples 1% of enterprise AI logs monthly for policy violations.
8. Violations → SEC review; repeat → access restriction per SOP-012.

Definition of Done (ongoing)

• [ ] Developer acknowledged AI policy annually
• [ ] No open SEC violations on AI usage

Related SOPs

• SOP-004 · SOP-011 · SOP-012