Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicWhat naming and tagging standard should AI agents follow?
Seven required tag keys for AI agents, with allowed values that fit AWS, Azure and Google Cloud tag and label limits, plus naming rules and enforcement.
How do you retire an AI agent safely?
An 11-step checklist to retire an AI agent: map dependants, stop traffic, revoke credentials, unlink tools and MCP servers, preserve evidence, watch stragglers.
What is an AI agent owner responsible for?
The duties of an AI agent owner as a RACI-style list with cadence: registration, access, evals, behaviour, cost, changes, evidence, succession and retirement.
When does a change to an AI agent need re-approval?
Which AI agent changes need re-approval: material, minor and emergency change classes with examples, triggers and the evidence each one requires.
What are AI agent autonomy levels?
AI agent autonomy levels describe how much an agent decides without a human, from operator to observer. A five-level scale and the controls each level needs.
Can you use an LLM to turn written AI policies into policy as code?
A workflow for drafting AI agent policy as code with an LLM: clause to rule to test cases to human sign-off, with a worked example and failure modes.
How do you build one CI/CD process for AI agents built on different platforms?
Common promotion stages and gates for code and low-code agents, each platform's export or versioning mechanism, and policy-as-code checks in the pipeline.
How do you apply different AI agent policies to different business units?
Use a global baseline plus business-unit overlays scoped by IdP groups or tags. Worked example: internet access for sales agents, blocked for underwriting.
Discovery and shadow agents
All in this topicHow do you find AI agents using cloud audit logs?
The CloudTrail, Azure Activity Log and Google Cloud Audit Log events that reveal AI agents, which are off by default, and how to turn hits into an inventory.
How do you link AI agents to the data assets in your data catalogue?
Join keys between your AI agent inventory and data catalogue, the lineage sources to use, and a view of which agents touch classified datasets.
Agent identity and access
All in this topicShould AI agents have their own accounts in SaaS apps?
User-delegated access vs shared service account vs dedicated agent identity in SaaS apps: a comparison on audit, least privilege, revocation and licences.
Can AI agents use just-in-time access?
Just-in-time access for AI agents: short-lived tokens, session policies, token exchange and approval-gated elevation, with real AWS and Google Cloud lifetimes.
How do you control OAuth consent for AI apps and agents?
Control OAuth consent for AI apps: Entra and Google Workspace settings, a four-tier scope risk model, an admin consent workflow and grant reviews.
What Okta permissions does an AI governance tool need to map AI usage to employees?
Least-privilege Okta setup for mapping AI usage to employees: an API service app, read scopes, a read-only admin role and what each scope exposes.
Runtime control
All in this topicHow do you roll out and roll back a new AI agent version safely?
A staged canary plan for AI agent versions: traffic percentages, evaluation gates per stage, rollback triggers and what to record for each release.
What is the difference between guardrails and evaluations for AI agents?
Guardrails block or change agent behaviour in real time; evaluations score it. Compare timing, cost, failure modes and the evidence each one produces.
Should an LLM decide how severe an AI agent incident is?
LLM-assigned vs rule-assigned severity for AI agent incidents: where model triage helps, the rule-based floors to keep, and how to audit severity changes.
How do you tell users when an AI agent is shut down or quarantined?
Who to notify when an AI agent is quarantined or killed, what fallback to give users, a message template, and how to wire alerts to the quarantine event.
Tools and MCP
All in this topicWhat is the difference between MCP and A2A?
MCP connects an agent to tools and data; A2A lets one agent delegate work to another. Compare purpose, discovery, transports, auth and what to govern.
Should you expose internal AI agents as MCP servers, A2A agents or REST APIs?
Choose MCP, A2A or REST for internal agents by consumer type, and see what each means for authentication, discovery, logging and versioning behind an API gateway.
How do you monitor MCP tool call failures across AI agents?
Log every MCP tool call with server, tool, argument hash, latency and error class; separate JSON-RPC errors from isError results; set per-server alerts.
Gateways and cost
All in this topicHow much latency does an LLM gateway add?
Where LLM gateway latency comes from, how to measure p95 overhead fairly, and which features cost the most milliseconds, with scoped benchmark figures.
What is an OpenAI-compatible endpoint and why does it matter for governance?
An OpenAI-compatible endpoint accepts OpenAI API-shaped requests, so apps switch providers by changing base_url. Why it helps governance and what differs.
How do you make an LLM gateway highly available?
Run stateless gateway replicas behind a load balancer, health-check them, fail over across regions and model providers, and size from your own load tests.
How do you track OpenAI API costs by project with the Usage and Costs APIs?
Use an admin key to pull daily costs by project and line item from the Costs API, tokens by API key and model from the Usage API, and give each agent a project.
How do you attribute AI spend to teams when your identity provider has no cost-centre data?
Fallback sources for AI cost attribution when the IdP lacks cost centres: HR system, manager chain, code-host org, cloud tags, agent owner, with confidence flags.
How do you allocate shared AI platform costs to divisions?
Split AI costs into direct and shared pools, allocate shared costs by usage, fixed split or headcount, and send each division a clear monthly statement.
Evidence and frameworks
All in this topicWhat does NYDFS AI cybersecurity guidance mean for AI agents?
How the NYDFS October 2024 AI industry letter maps to 23 NYCRR Part 500 controls for AI agents, with the evidence to keep for each section.
What is an AI bill of materials (AI-BOM)?
What an AI-BOM is, what CycloneDX ML-BOM and SPDX 3.0 AI profile fields define, and what an AI agent's AI-BOM must add: tools, MCP servers and skills.
What does Singapore's agentic AI governance guidance expect from enterprises?
Singapore has IMDA's Model AI Governance Framework for Agentic AI (updated May 2026) and MAS AI risk guidelines for banks, effective October 2027.
Buying and comparison
All in this topicHow do you avoid lock-in when choosing an AI governance platform?
A 10-point checklist with pass/fail tests: open telemetry, standard identity, MCP and OpenAI-compatible APIs, full data export, and no agent migration.
What should acceptance testing cover for an AI agent governance platform?
Twelve acceptance tests with pass criteria for an AI agent governance platform: discovery, ownership, traffic paths, redaction, SSO, cost and quarantine.
What role should a systems integrator play in an AI agent governance rollout?
A RACI for vendor engineers, systems integrators and internal teams in an AI agent governance rollout, with per-platform work and SI exit criteria.
Why does an AI agent governance platform cost more than an LLM gateway?
A like-for-like scope comparison of an LLM gateway and an AI agent governance platform, plus the hidden costs of running a gateway yourself.
Should you buy AI agent governance as separate modules or as one suite?
When to buy AI agent governance as point modules (like an LLM gateway) vs one suite: what you own, data-model stitching costs, and a crawl-walk-run plan.
Platform Guides
All in this topicHow do you control ChatGPT Atlas and agent mode in the enterprise?
Atlas is off by default for ChatGPT Enterprise. Admins control sign-in, agent mode and memories via RBAC, but Atlas emits no Compliance API logs. What to know.
How does Databricks govern managed and external MCP servers?
Databricks governs MCP servers through Unity Catalog and Unity Gateway: grants decide who can call each server, policies govern calls. Some endpoints bypass it.
Can Cursor hooks block risky agent commands?
Yes. Cursor hooks such as beforeShellExecution and beforeMCPExecution can allow, ask or deny agent actions. Here is how to deploy them and avoid fail-open gaps.
What is Okta Cross App Access, and how does it control AI agent access?
Okta Cross App Access (XAA) lets your IdP decide which AI apps and agents may reach which enterprise apps, using ID-JAG tokens instead of per-app consent.
What does the ChatGPT Enterprise Compliance API let you audit?
The Compliance API exports ChatGPT Enterprise conversations, files, GPT configs, memories and users for eDiscovery and DLP. Here is what it misses.
Can Claude Code hooks enforce security policy across an enterprise?
Partly. Managed PreToolUse hooks can block risky Claude Code actions org-wide, but they run on the device, fail open by default and need backing controls.
How do you restrict GitHub Copilot coding agent's internet access?
Use the built-in agent firewall: set it at organisation level, decide on the recommended allowlist, add narrow custom entries, and know what it does not cover.
How do you audit Agentforce agent sessions with Session Tracing?
Agentforce Session Tracing writes sessions, turns, steps and messages to Data 360 objects you can query, giving a step-level audit trail of each agent.
How do you tag Amazon Bedrock costs by agent with application inference profiles?
Create one application inference profile per agent, tag it, activate the tags in Billing, and make every call use the profile ARN, or the tags never apply.
How does agent identity work in Vertex AI Agent Engine?
Agent identity gives each Agent Engine agent its own SPIFFE-based IAM principal, so you grant roles per agent instead of to a shared service account.
How does Amazon Bedrock AgentCore Identity manage OAuth tokens for agents?
AgentCore Identity keeps agent OAuth tokens and API keys in a token vault, issues workload access tokens, and checks inbound callers with SigV4 or JWT.
Can Conditional Access policies apply to AI agents in Microsoft Entra?
Yes, with limits. Entra Conditional Access can target agent identities and agent users, block on agent risk, and lock agent users to compliant devices.
What does Microsoft Agent 365 govern, and what does it leave to other tools?
What Microsoft Agent 365 covers since its May 2026 GA, how its read-only registry sync for other platforms works, and which controls need another layer.
How do you stop Copilot Studio agents from running on the maker's credentials?
How the Copilot Studio maker credential control works, why autonomous agents fail when you enforce it, and how to find agents still using a maker's connection.
How do you govern AI agents built in Cloudera Agent Studio?
What Cloudera secures in Agent Studio, what its shared responsibility model leaves to you (MCP allow-lists, token limits, secrets), and how to close the gaps.
Put the framework into practice