Classification Memos & Sign-off
Generated by KlarvoEngine: Classification memos are auto-generated by KlarvoEngine after the 3-pass classification pipeline completes. The memo includes article citations, risk level reasoning, and applicable obligations. See Understanding Your Classification Memo.
A Classification Memo is the auditable document that records how and why an AI system was classified at a particular risk level. It's auto-generated after completing KlarvoEngine classification and serves as the foundation of your compliance evidence for that system.
What a Classification Memo Contains
Every memo includes:
| System Identity | Name, internal ID, department, vendor (if applicable) |
| Classification Result | Risk level: Minimal / Limited / High-Risk Candidate / Blocked |
| Confidence Level | High / Medium / Low — based on answer certainty |
| AI Definition Test Result | Likely AI System / Likely Not / Needs Review + rationale |
| Prohibited Screening Result | No indicators / Potential prohibited (with details) / Needs legal review |
| High-Risk Screening Result | Matched categories (if any) + rationale |
| Transparency Screening Result | Applicable scenarios (if any) |
| Applicable Obligations | List of obligations triggered by the classification |
| Reviewer | Name of the person who reviewed and confirmed |
| Sign-off Date | Date of formal approval |
| Classification Rationale | Free-text explanation of the reasoning |
The Sign-off Process
Classification requires formal reviewer sign-off:
Complete the wizard: All classification steps must be answered
Review pre-filled classification: The engine suggests a classification based on your answers
Override if needed: If you disagree with the engine's suggestion, you can override — but must provide a written justification (recorded in audit trail)
Assign reviewer: Select a team member with Reviewer/Approver or higher role
Confirm sign-off: Enter the sign-off date and any notes
Generate memo: The PDF is auto-generated and stored
Human Override
The classification engine provides a suggestion, but humans always have the final say:
If you override the engine's suggestion, you must provide a reason
The override is permanently recorded: original engine result, override decision, and rationale
Overrides are flagged in the audit trail for transparency
This is a feature, not a bug — the engine can't capture every contextual nuance
Version History
Classifications can change over time. Every change creates a new version:
Original classification: V1 with initial wizard answers
Re-classification: V2+ with change reason documented
Reassessment-triggered: When material changes, incidents, or vendor updates occur
Each version is preserved — you can always see the full classification history, which demonstrates ongoing compliance.
Downloading and Sharing
To download a Classification Memo:
Open the AI System detail page
Click Exports → Classification Memo
Choose PDF format
The memo downloads with your organization branding, document control, and confidentiality marking
You can also include the memo in a full Evidence Pack ZIP export.
Best Practices
✅ Always sign off: An unsigned classification has limited audit value
📝 Write clear rationale: A reviewer should understand your reasoning without asking questions
🔄 Re-classify when things change: New features, new user groups, vendor changes — all merit re-classification
📋 Store in evidence vault: The memo PDF is automatically linked; ensure related evidence is also attached