Audit log

Tamper-evident record of every model input, output, version, and reviewer action.

An audit log captures the full provenance of an AI decision: input, retrieval results, prompt version, model version, output, confidence score, reviewer action, and final decision. This is non-negotiable for risk-compliance engagements; the audit log is what we hand to risk officers when they ask 'why did the system do this?'

When it matters

Always for production AI. Required for compliance, debugging, and post-incident analysis. The single most under-built piece of AI infrastructure in 2026.

Real example

Every inference call writes: timestamp, user_id, prompt_version, model_id, input_hash, output_hash, tool_calls, reviewer_action, latency. Queryable from the same dashboard as KPIs. Used last quarter to trace a quality regression to a prompt change in under 15 min.

KPIs to watch

Audit log completeness (100% required), log retention (12+ months for regulated workflows), query latency on log analytics (<5s for typical questions).

Related terms

AI governance

Policies, processes, and controls that make an AI system auditable and accountable.

Model card

Documentation describing a model's intended use, limitations, evaluation, and risks.

Grounding

Anchoring model output to verifiable source material to reduce hallucination.

Hallucination

Plausible but factually incorrect output generated by an LLM with no grounding.

See it in action

We use this every week

Book a 30-min call and we'll walk you through how Audit log shows up in a real engagement we're running.

Book a 30-min call