Audit Log

Audit log statistics show that among the four models used, two models produced hundreds of real operations through the harness, while the other two produced zero real operations.

Common means of mitigating the permission risks brought by Agents autonomously opening accounts include using short-lived credentials, limiting scope by task, requiring manual secondary confirmation for sensitive operations, and recording complete invocation audit logs.

Traditional audit logs are a post-hoc mechanism and cannot intercept or approve actions before they occur

Immutable storage or cryptographic signatures should be applied to critical operation logs to ensure they cannot be silently modified after the fact.

Every operation an Agent performs on an external system should be fully logged to form structured audit records, enabling traceability of the operation's initiator, timestamp, context, and specific changes.

Audit log retention is generally required to be at least 1 year

When the fire-and-forget pattern is applied to audit logging scenarios, messages may be lost due to network partitions, queue overflow, consumer crashes, and other reasons, with the caller completely unaware of this.

When an operation is successfully executed but the audit log write fails, it leaves a compliance gap with no alerts of any kind.

Including audit logs in the same transaction increases lock-holding time and write volume for the transaction, which can become a performance bottleneck in high-concurrency scenarios.

The intelligent proxy layer can fully record every operation initiated by the AI (HTTP calls, command executions, resource accesses), forming a traceable audit log.