Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicWhat should an AI acceptable use policy say about agents?
A clause-by-clause checklist for the agent section of an AI acceptable use policy: building, registering, connecting tools, data classes, approvals and disclosure.
What are the stages of the AI agent lifecycle?
Seven AI agent lifecycle stages from intake to retirement, with entry and exit criteria and the governance artefact each stage should produce.
What is an AI agent registry?
An AI agent registry is the system of record for agents, versions, owners, tools and permissions. How it differs from a CMDB and a model registry.
Which metrics show your AI agent governance is working?
Ten AI agent governance KPIs with formulas and starting targets: inventory coverage, ownership, policy enforcement, eval coverage, time to quarantine and cost.
How do you risk-tier AI agents?
A three-tier rubric for AI agents: score actions, autonomy, data, reach and regulated use, apply overrides, then attach required controls to each tier.
How is AI agent governance different from AI model governance?
Model governance checks outputs; agent governance controls actions. Eight agent-only control areas, and why 2026 US bank guidance left agentic AI out of scope.
Discovery and shadow agents
All in this topicHow do you find AI agents in your code repositories?
Find AI agents in your repositories by searching dependency manifests, SDK imports, MCP config files and agent instruction files. Includes ready-to-use queries.
How do you detect unapproved LLM API traffic on your network?
Find unapproved LLM API traffic using a provider domain list, DNS and secure web gateway logs, TLS SNI and cloud flow logs, and handle the false positives.
How do you find AI features switched on inside SaaS apps?
Find embedded AI in SaaS apps: review admin consoles, contracts and sub-processor lists, and OAuth grants in your identity provider. A step-by-step method.
How do you find leaked or unused LLM API keys?
Find leaked and unused LLM API keys: check last-used age, owner and scope, use secret scanning for provider key patterns, and rotate on a schedule.
Agent identity and access
All in this topicRuntime control
All in this topicHow do you validate an AI agent's output before it acts?
Validate AI agent output in sequence: schema checks, business rules, groundedness and task adherence, then risk-based approval, before any tool call executes.
How do you control which websites and APIs an AI agent can reach?
Control AI agent network egress with per-agent allow-lists at an egress proxy, DNS policy and network identity, plus tool-level limits. A step-by-step guide.
How do you stop an AI agent from looping?
Stop AI agent loops with a max-step limit, a per-task token budget, repeated tool-call detection and a cost circuit breaker enforced outside the agent.
How do you detect behavioural drift in AI agents?
Detect AI agent behavioural drift by baselining tool-call mix, tokens, refusals and errors per agent version, then alerting on sustained deviation.
What should an AI agent incident response runbook include?
A phase-by-phase AI agent incident runbook: detect, contain, preserve evidence, investigate, recover and learn, with timings, roles and artefacts to keep.
What is excessive agency in AI agents?
Excessive agency (OWASP LLM06) is when an agent can do more than its task needs. The three root causes, examples and one control for each.
How do you stop memory poisoning in AI agents?
Memory poisoning plants instructions in an agent's long-term memory or context. Controls for writes, provenance, expiry and detection, mapped to OWASP ASI06.
What is policy as code for AI agents?
Policy as code for AI agents: rules written as versioned, tested files that decide what agents may call, send and spend, with examples and a review workflow.
Gateways and cost
All in this topicHow do you rate-limit AI agents?
Rate-limit AI agents with per-agent token and request quotas, burst allowances and priority tiers, so one agent cannot drain the shared provider quota.
What is LLM model routing?
LLM model routing sends each request to the model best suited by rule, cost, availability or difficulty. Four strategies and when each pays off.
How do you measure cost per task for an AI agent?
Cost per task = all model, retry, tool and human-escalation costs for an agent, divided by tasks completed successfully. Worked example inside.
Can Azure API Management act as an AI gateway for agents?
Azure API Management has AI gateway policies for token limits, token metrics, semantic caching and content safety, plus MCP support. What it covers for agents.
How should users and applications authenticate to an LLM gateway?
Separate console login from machine access: SSO for people, and scoped virtual keys, OAuth client credentials or workload identity for apps.
Evidence and frameworks
All in this topicWhat evidence do AI agents handling PHI need under HIPAA?
A checklist of HIPAA evidence for AI agents that touch PHI: risk analysis, access control, audit controls, minimum necessary, transmission security and BAAs.
What goes into an AI agent card?
An A2A Agent Card lists an agent's name, interfaces, capabilities, skills and security schemes. See which governance fields auditors also need and where to keep them.
How do you run an AI impact assessment for an AI agent?
Apply the ISO/IEC 42005:2025 impact assessment structure to an AI agent, section by section, with the agent-specific questions on tools, autonomy and delegation.
What should an AI agent audit log contain?
A field-by-field checklist for AI agent audit logs, aligned to OpenTelemetry GenAI attributes, plus the on-behalf-of actor and policy decision fields auditors ask for.
ISO/IEC 42001 vs NIST AI RMF: which should you use for AI agents?
ISO/IEC 42001 is a certifiable management system; NIST AI RMF is a voluntary risk framework. How they differ, and one crosswalk for agent controls.
Do AI agents have to disclose they are AI under the EU AI Act?
Yes. EU AI Act Article 50 has applied since 2 August 2026: agents that interact with people must say they are AI unless it is obvious. What to check.
Buying and comparison
All in this topicHow long does it take to roll out AI agent governance?
AI agent governance rolls out in phases: read-only discovery first, then registry, observation and inline enforcement. What slows each phase down.
What is the difference between AI-SPM and an AI agent control plane?
AI-SPM finds risky AI configurations; an agent control plane also governs agent lifecycle and enforces policy at runtime. Capability table and buyer tests.
What questions reveal whether an AI agent discovery tool really covers your estate?
Test AI agent discovery with planted agents, keys and MCP servers across clouds, SaaS and laptops. Questions to ask, scoring metrics and red flags.
Platform Guides
All in this topicHow do you govern AI agents built in Dify?
Dify governance: workspace roles, model provider keys, plugin permissions and signature checks, Enterprise install policy, and routing model calls via a gateway.
How do you keep LLM and agent traffic off the public internet on AWS, Azure and Google Cloud?
Private model access on Bedrock (PrivateLink), Microsoft Foundry (private endpoints) and Vertex AI (PSC, VPC-SC), plus the agent egress paths they miss.
How do you govern Kiro and Amazon Q Developer in the enterprise?
Kiro and Q Developer governance: enterprise sign-in, profile-level MCP toggle and registry allow-list, model approval, activity reports and S3 prompt logs.
How do you monitor Claude Code usage with OpenTelemetry?
Claude Code OpenTelemetry setup: env vars, metrics and events, managed settings, cost attribution by team, cardinality switches and which content to log.
How do you control which models GitHub Copilot can use?
Copilot Business and Enterprise owners enable or disable each model, set new-feature defaults and cap spend with budgets. How policies resolve across orgs.
What does GitHub Copilot content exclusion protect, and where does it stop?
Copilot content exclusion blocks suggestions and context from excluded files, but not in VS Code agent mode, symlinks or remote filesystems. Rules, limits, fixes.
What does Snowflake Cortex Guard filter, and is it enough for agents?
Snowflake Cortex Guard filters harmful COMPLETE responses. What it catches, how it is billed, how to monitor it, and the agent risks it does not cover.
How do you govern agents built with Pydantic AI?
Pydantic AI gives typed outputs, OpenTelemetry tracing, usage limits and tool approval. How to use them for governance and what to add around them.
How do Power Platform data policies control Copilot Studio agents?
Power Platform data policies block Copilot Studio connectors, channels and unauthenticated agents. How enforcement works and a starting policy to copy.
How do you test Salesforce Agentforce agents before deployment?
How Agentforce Testing Center checks topics, actions and outcomes, how to build test sets, and how to keep results as release evidence.
How do you govern model access through SAP generative AI hub?
SAP generative AI hub orchestration offers content filtering, data masking and grounding modules. What each control does and what to add for agents.
How do you govern agents built with OpenAI AgentKit?
What Agent Builder, Connector Registry, Guardrails and Evals give you for governance in OpenAI AgentKit, and the controls you still need around them.
How does Databricks Unity Catalog govern tools that agents call?
Unity Catalog functions become agent tools governed by USE CATALOG, USE SCHEMA and EXECUTE grants. Setup, permissions checklist, audit and gaps.
How does Policy in Amazon Bedrock AgentCore control agent tool calls?
Policy in Amazon Bedrock AgentCore checks every agent tool call at the Gateway against Cedar rules. How it works, default deny, LOG_ONLY mode and its limits.
What is Amazon Bedrock AgentCore Gateway and how do you govern it?
AgentCore Gateway turns APIs, Lambda functions and MCP servers into one MCP endpoint. Its target types, inbound and outbound auth, and what to govern.
What does Microsoft Purview DSPM for AI cover for AI agents?
What Purview DSPM for AI sees across Copilot, Copilot Studio agents, enterprise AI apps and third-party AI sites, its prerequisites and where it stops.
How do you govern kagent agents on Kubernetes?
kagent defines agents, models and MCP tools as Kubernetes resources. Govern them with RBAC, GitOps review, tool allow-lists and OpenTelemetry.
Put the framework into practice