Auditing AI Agent Usage

On This Page

Appian provides system-level audit logs for AI capabilities, including AI agents. These logs help you meet security and compliance requirements by tracking AI interactions across your environment.

This page explains the two types of AI audit logs and their limitations. These logs are designed for platform-level security and compliance auditing, which is different from the Monitor tab in the AI agent screen. The Monitor tab helps developers track performance and logic for individual AI agent runs. For monitoring guidance, see Monitor AI Agents.

AI audit log types

Appian provides two system-level audit logs for AI capabilities:

AI Usage Audit Log: Tracks all AI interactions by default, providing a high-level audit trail for compliance.
AI Input/Output Log: Optionally logs sensitive data, including prompts and responses. This log is off by default and must be enabled by an administrator.
AI Guardrail Violation Log: Records metadata for every guardrail violation, including timestamp, user, guardrail, action taken, and source object. Always active when guardrails are enabled. Located at audit/ai_guardrail_violations.csv.

AI Usage Audit Log

The AI Usage Audit Log tracks all AI interactions across your Appian environment by default. This log provides the audit trail required for security questionnaires and compliance reviews.

The AI Usage Audit Log is always enabled and requires no configuration.

Note: Process model node suggestions are not captured in the AI usage log.

For complete log structure details, see AI usage logs.

AI Input/Output Log

The AI Input/Output Log captures sensitive data, including the actual prompts sent to the AI and the responses received. This log is off by default to protect sensitive information.

The AI Input/Output Log is accessible only by system administrators.

Note: The following features are not captured in the AI input/output log:

Process model node suggestions
Inline testing in the AI skill designer
Smart search data

For complete log structure details, see AI input/output logs.

Enable input/output logging

To enable the AI Input/Output Log:

In the Admin Console, go to AI.
Under AI Input/Output Logging, select Enable logging of AI inputs and outputs.
Click Save Changes.

Caution: The platform does not mask personally identifiable information (PII) or other sensitive data in these logs. Only enable this log if your security and compliance requirements allow logging of unmasked data.

Log retention and limitations

AI audit logs have specific retention policies and limitations:

Logs are retained for a maximum of 7 days.
The cumulative size of all AI logs cannot exceed 100 MB.
When the size limit is reached, the oldest logs are automatically removed.
Only system administrators can access the AI Input/Output Log.

MCP tool calls

When an AI agent calls an MCP tool, the tool call is recorded as part of the AI agent run. MCP tool calls appear in the Monitor tab step history alongside native tool calls. MCP tool calls are also captured in the audit logs.

For more information about AI agents:

For developer-focused monitoring and performance tracking, see Monitor AI Agents.
For security considerations, see AI Agent Reference.
For detailed log structure and metrics, see AI input/output logs and AI usage logs.

Feedback

Was this page helpful?