AI Governance v.41 release - Review what's new, fixes and limitations in AI Governance for the v.41 release.

Automation 360
Review what's new, fixes and limitations in AI Governance for the v.41 release.

What's new

The AI Governance landing page now provides a centralized audit log of all AI Agent activity and interactions with LLM models.
AI Agent Logs gives governance, risk, and compliance (GRC) and admin users a real-time audit trail of every AI Agent activity, including interaction with LLM models and tool calls for compliance and responsible AI oversight. See, AI Agent Governance Logs.
  • The landing page summarizes status details, offering quick insight into execution status.
  • Each agent run, from trigger to completion, is captured as a single record with a full breakdown.
  • Users can quickly navigate to activity details for each run, from the landing page.
  • The Detail view captures tool calls, inputs and outputs, toxicity scores, and token usage across the full agent execution lifecycle.
Tool calls can be audited now with support from AI Governance logs.

Agent Interoperability extends AI Governance logs to capture all third-party agent tool calls, including MCP inbound calls. Automation experts now have a traceable audit trail of every cross-agent transaction.
Upload datasets to compare agents in AI Evaluations.

Evaluation Datasets enables pro developers to supply test inputs and expected outputs for AI Agent evaluations. Datasets can be uploaded as JSON, selected from an existing library, or entered manually. This gives teams a flexible, repeatable way to batch-execute agents and score performance against defined benchmarks.

Manual evaluations can now be performed on AI Agents.

Evaluators can now manually score AI Agent outputs using human judgment alongside automatic evaluations. Reviewers assign Pass/Fail ratings per test case (through thumb up/down), comparing actual outputs against expected results. Aggregated scores are surfaced in the Evaluation Summary view, with all feedback logged for trend tracking and continuous improvement.
Summaries are now available for AI Evaluations performed manually on skills and agents.

The Evaluation Summary gives pro developers and evaluators a centralized view of human-scored evaluations for both AI Agents and AI Skills, aggregating the manual Pass/Fail judgments to validate performance and support continuous model improvement.

Fixes

GEN_AI_AGENT returned to Source Filter

The GEN_AI_AGENT option now appears in the Source filter, as expected. Previously, the GEN_AI_AGENT source option was missing from the Source filter in the AI Prompt and Event logs.

Limitations

In AI Governance logs, the agent steps output data displays raw JSON, instead of structured, human-readable rows.
The AI guardrail masking feature does not mask the response from task bots in AI Skills. This is being addressed. See, Data masking in AI.
Workaround: In the AI Skill, include the following user prompt. 
Please keep all HTML Tag prefix and suffix unchanged in both the request and the response.