Agent monitoring and insights plus security metrics in watsonx.governance

Risk and compliance leaders, security teams and AI owners often ask these two questions: “Can I see exactly what my AI agents are doing in production?” “Can I get a unified view of my AI Governance and AI security posture?”

With watsonx.governance, the answer to both is now yes.

1. Agent monitoring and insights: Giving enterprises visibility into agentic AI behavior, actions, and decisions in production.
2. Security metrics in the governance console: Bringing Guardium AI Security insights directly into your governance workflows.

Together, these features will deliver a comprehensive solution for AI trust—policy, protection, and performance—without spreadsheet chasing or tab-hopping.

Enterprises are turning to AI agents as the next frontier of productivity. Unlike traditional models that simply generate outputs, agents can take action—chaining tasks and tapping into multiple systems. This opens the door to automating repetitive workflows, accelerating decision-making, and freeing teams to focus on higher-value work.

But with this promise comes new risks. Agents operate with more autonomy, making it harder to see how decisions are made and to ensure they’re behaving as intended. Monitoring today is often slow, manual, and fragmented, leaving developers without the tools they need to trust, track, or fine-tune their systems. And without a scalable governance infrastructure, enterprises struggle to evaluate, control, and confidently scale agentic AI.

To combat this challenge, in our upcoming releases, watsonx.governance will introduce Agent Monitoring and Insights in watsonx.governance. This new capability monitors agentic applications in production. By tracking decisions, behaviors, and performance in real time, Agent Insights issues alerts when metrics cross thresholds. This enables proactive management, faster troubleshooting, and higher confidence in agent-driven outcomes.

• In the loop evals: Updated metrics to accurately assess the performance (quality and behavior) of AI agents. This saves time by increasing efficiency by viewing all performance metrics aggregated at Conversation, Interaction and Tool level in one place, painting a complete picture of agent performance. 
• Drive quality and precision: In Q1 2026, we are introducing actionable insights via root cause analysis. This will help teams have the ability to measure what’s working, compare performance over time, and continuously optimize and fine-tune their AI and agents.
• Real-time experimentation tracking: Evaluate and compare multiple runs of your agentic applications, with all changes, metrics and outputs in one place. This not only accelerates time to production—gaining structure and visibility to the agent development process with all changes, metrics and outputs—but it also improve efficiency, which cuts through noise by focusing only on the metrics that matter most to your project through custom rankings tailored to your project’s success metrics — ensuring faster decisions and fine-tuned agent performance.
• Reports: Built-in dashboards and automated alerts when AI agents go outside of the bounds of intended operation,  to help teams test, adopt and scale agents faster. This ensures proactive management, which drives proactive management with the Agent Insights Dashboard, and tracks performance trends over time. It also mitigates risk in real-time, gaining confidence by ensuring you’re promptly notified if an agent makes an error or hallucinates, enabling timely intervention and correction.

• Risk and compliance gain assurance that agents stay within approved scope and decisions are auditable.
• AI/ML Ops and developers troubleshoot faster, improve efficiency and ensure reliable outcomes.
• Business leaders get higher confidence to scale agentic AI programs responsibly to extract maximum ROI.

An AI agent automating procurement might attempt to approve a vendor contract outside its scope. With monitoring enabled, that action is flagged in real time, allowing teams to investigate, adjust policies, or refine the agent before it causes downstream issues.

AI governance and AI security have too often been treated as parallel workstreams. That changes with the new integration of Guardium AI Security into watsonx.governance console. Risk and compliance leaders can view live security posture directly where they approve and manage use cases.

• Per–use case security panel: Each AI use case surfaces a concise security card: vulnerability scan and pen-test results, real-time detections (for example, prompt injection attempts) and 7/30-day trends.
• Program-level security dashboard: Aggregate view across use cases, open findings by severity, attack activity over time and remediation status.
• Roadmap of Expanding Coverage: Day-one metrics focus on vulnerabilities, pen-tests and detections, with misconfigurations and more signals coming soon.

• Risk and compliance accelerate reviews with live security data providing real-time visibility to critical vulnerabilities.
• Security teams using Guardium for deep discovery and protection, can easily share key metrics with AI governance stakeholders. AI owners see actionable metrics tied to their use cases, providing a complete view of security vulnerabilities and risks that must be remediated.

An AI agent that manages IT tickets is registered and risk-tiered in governance. Guardium continuously tests it for abuse or leakage. Now, high-severity findings, pen-test dates and blocked attempts appear directly next to the risk record—so approvals happen faster with full context.

Both new features are guided by the same principle: your AI program should run on one unified source of the truth. Agent monitoring will evolve with a governed agent catalog, fine-tuning controls and richer observability to ensure AI agents remain reliable, accountable and aligned.

Security metrics will continue to expand beyond vulnerabilities and detections into misconfigurations and deeper Guardium integrations.

Together, they will bring security and governance into a unified flow so enterprises can govern and scale AI with speed, safety and confidence.

If you already use watsonx.governance, look out for the release and be ready to activate these features on your most critical AI use cases as they become available.

 Your IBM team can help you identify the right path forward and prepare to build governed, secure, and trustworthy AI with these upcoming capabilities.

Scale trusted AI with watsonx.governance