Why AI audit logs break when the facts change

Short summary

AI agent audit logs typically capture prompts and tool calls but fail at decision reconstruction when underlying facts change over time. The article argues for distinguishing event time, valid time, and system time, and preserving immutable source versions alongside corrections. The author's startup Lians is building a system of record for AI in regulated workflows and seeks design partners.

• •AI audit logs must separate event time, valid time, and system time to reconstruct past decisions
• •Logging only IDs is insufficient if documents or policies behind them can change later
• •Author's startup Lians seeks 5-7 design partners in financial research, risk, and compliance

Generated with AI, which can make mistakes.