AI Proof
AI as an auditable assistant, not an invisible approver.
TwentyCore AI should help users analyze, draft, summarize, and prioritize. Posting, approval, submission, and overrides stay controlled by users and configured workflows.
Principles
The buyer-facing AI control model.
Tenant-scoped context
AI prompts should use only the active tenant data needed for the task.
Human confirmation
AI can draft, explain, and summarize, but sensitive actions require user acceptance.
Minimum necessary data
Prompt context should avoid unrelated records and sensitive payloads.
Provider disclosure
AI provider, retention, and training terms must be confirmed per deployment.
Demo prompts
Prompts tied to the seeded demo tenant.
- Summarize open sales orders and production risk for this tenant.
- Which customer orders should I review first this week?
- Explain why PRD-2026-0001 may affect SO-2026-0201.
- Draft a follow-up note to Customer A about delivery status.
Evidence
What reviewers can ask to see.
- tasks/artifacts/ai_endpoint_audit_2026-04-29.md
- backend/tests/regression/test_ai_tenant_context.py
- backend/tests/regression/test_ai_prompt_no_hardcoding.py
- docs/sales/ai-proof-methodology.md