What it is
An audit trail is a log of every action taken by or through an AI agent, with decision metadata. Each entry captures: the agent identity, the action attempted (tool call), the policy verdict (allow/deny/escalate), the threat category if denied, the severity score, the timestamp, the LLM model used, and the human reviewer decision (if escalated).
Critically, the audit trail is immutable and append-only. New entries can be added, but existing entries cannot be modified or deleted. This is typically enforced by writing to a system the agent process doesn't control: a dedicated audit log service, a write-once storage system, or an external compliance database.
Why it matters
Two reasons: forensics and compliance. If something goes wrong, the audit trail provides a complete record of exactly what the agent attempted, in what order, and what policy decisions were made. This is essential for incident response: did the agent attempt the deletion or was it the human operator?
For compliance, regulators ask: prove that you have controls in place. SOC2 Type II audits require "continuous monitoring evidence" — a trail showing that every action was evaluated. ISO 27001 requires "access control evidence" — proof that only authorized subjects accessed protected resources. The audit trail is that evidence.
How it works
Every time the proxy makes a governance decision, the decision is logged: agent ID, action details, policy rule applied, decision, threat category, severity score, timestamp. The log entry is written to a system outside the agent's trust boundary. If the agent is compromised or misbehaving, it cannot modify the audit log it doesn't control.
The logs are queryable and exportable. For forensics, you might ask: "What did agent-trading-bot attempt between 3pm and 4pm?" For compliance, you export the full month's log and submit it to an auditor.
How Intercis implements it
Intercis maintains an immutable audit log for every proxy interception event. Each log entry includes: agent ID, tool call content, policy verdict, threat category, severity score, timestamp, LLM model, human decision (if any). The log is tamper-resistant — the agent process cannot modify entries it didn't write (and isn't supposed to write any).
We support CSV export for compliance workflows, integration with SIEM systems, and queryable dashboards. The log can be exported on-demand or on a schedule for SOC2 audit submissions.
Related terms
- AI Agent Governance — The broader discipline the audit trail supports.
- Human-in-the-Loop — Escalated decisions are also logged in the audit trail.