Watsonx Orchestrate Plug‑ins for Security and Compliance

2026-03-10 · Source: Niklas Heidloff · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Intermediate, short

Summary

IBM watsonx Orchestrate agents now support plug-ins, enabling customization of how agents process inputs and generate outputs. These plug-ins are crucial for enforcing safety, security, and regulatory compliance by filtering, sanitizing, or transforming data. There are two types: input plug-ins, which operate before an agent processes a request (e.g., masking "silly" or blocking "stupid"), and output plug-ins, which refine or change the final response (e.g., masking email addresses). Examples demonstrate a Python-based `guardrail_plugin` for input moderation and an `email_masking_plugin` for output sanitization. Plug-ins are registered within an agent's definition, such as an `email_agent` utilizing `watsonx/meta-llama/llama-3-2-90b-vision-instruct`.

Key takeaway

For AI Engineers building agents with watsonx Orchestrate, integrating plug-ins is essential for robust compliance and security. You should define both input and output plug-ins to establish guardrails, ensuring sensitive data is handled appropriately and agent responses adhere to organizational policies. This proactive approach minimizes risks associated with unmoderated AI interactions.

Key insights

Watsonx Orchestrate plug-ins enhance agent reliability by customizing input/output for safety and compliance.

Principles

• Pre- and post-processing enhances agent control.
• Custom logic enforces guardrails and data privacy.

Method

Implement Python-based input plug-ins (AGENTPREINVOKE) to modify or block user requests, and output plug-ins (AGENTPOSTINVOKE) to refine agent responses, registering them in the agent's YAML definition.

In practice

• Mask sensitive words in user input.
• Block requests containing forbidden terms.
• Sanitize PII like email addresses in agent outputs.

Topics

• watsonx Orchestrate
• AI Agents
• Plug-in Architecture
• Data Guardrails
• Content Moderation

Best for: AI Engineer, MLOps Engineer, AI Security Engineer

Related on AIssential

• Stop AI Agents Before They Make Risky Moves — AI agents require a human-in-the-loop safety layer to mitigate risks from autonomous actions and ensure accountability.
• MCP Tools acting On‑Behalf‑Of Users in Orchestrate Agents — SSO and OAuth OBO flows in watsonx Orchestrate enable secure, seamless agent impersonation for automated task execution.
• Symbolic Guardrails for Domain-Specific Agents: Stronger Safety and Security Guarantees Without Sacrificing Utility — Symbolic guardrails offer strong, practical safety and security guarantees for AI agents without sacrificing utility.
• Safeguard your agentic AI applications with the Amazon Bedrock Guardrails InvokeGuardrailChecks API — The InvokeGuardrailChecks API offers resourceless, score-based safety checks for granular control in agentic AI workflows.
• Harnesses in AI: A Deep Dive — Tejas Kumar, IBM — AI harnesses ground non-deterministic models in reality, ensuring reliable agent behavior through structured control and verification.
• Importing LangGraph Agents in watsonx Orchestrate — IBM watsonx Orchestrate now supports importing LangGraph agents, enabling flexible, collaborative AI systems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Niklas Heidloff.