Audit Trail
Complete history of AI interactions with classified context
6
Sessions This Week
15
Files Accessed
262
Fields Allowed
53
Fields Blocked
All sessions permanently verified on Hedera Hashgraph (enterprise ledger)
Viewing as:
Different roles see different levels of detail
Showing 6 of 6 sessions
What Gets Logged
Transparency into our audit trail design
LOGGED
User & roleJennifer Walsh, M&A
∟ Accountability
TimestampDec 10, 10:15 AM
∟ Timeline
Duration23 minutes
∟ Usage patterns
Message count14 messages
∟ Activity level
Files accessedacme_financials.xlsx
∟ What was touched
Field counts3 allowed, 5 masked, 13 blocked
∟ Proof controls worked
AI modelClaude Sonnet 4
∟ Vendor tracking
Blockchain hash0x7a3f...b2c1
∟ Immutability
NEVER STORED
Session summaryWhat you discussed
∟ Leaks business context
Individual messagesYour questions
∟ Privacy, liability
AI responsesWhat AI said
∟ Privacy, liability
Actual field valuesThe $42M number
∟ Not needed for audit
Business contextWhy you asked
∟ Zero context retained
📋 What Auditors Actually Need
Auditors don't care about the conversation. They care about:
Who accessed what files
What was protected (field counts)
When it happened
Proof it's immutable (hash)
The actual messages? Not needed. In fact, storing them is a liability.
✨ Benefits of This Design
Privacy by Design
If someone breaches your audit logs, they get:
✓ "Jennifer Walsh analyzed M&A documents, 13 fields blocked"
✗ "What's the acquisition price?" → "Based on the CIM, $42M..."
Data Minimization
GDPR Art. 5 compliant — only storing what's necessary for compliance.
90% Less Storage
Full logs: ~50KB/session. Summary logs: ~2KB/session.
On-Premise AI = No Leakage
Summary generated on-prem. It describes interaction patterns, not content.
Simpler Legal Discovery
You have who, when, what was protected. No content = less liability.