Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicWhat risks do AI agents add beyond generative AI chatbots?
How tools, memory, autonomy and delegation create new failure modes for AI agents, mapped to OWASP agentic risk IDs and the control for each.
What is the difference between guardrails, policies and controls for AI agents?
Precise definitions of policies, controls and guardrails for AI agents, one example of each, and where each lives: model call, gateway, platform or process.
What should an internal AI agent governance policy contain?
A section-by-section template for an internal AI agent governance policy, with sample clause wording for scope, roles, risk tiers, runtime controls and retirement.
What is the minimum AI agent governance a company should have?
Six controls to put in place in the first 30 days of AI agent governance, each with a clear done criterion and the evidence it produces.
What does an AI agent governance maturity model look like?
Five maturity levels for AI agent governance, each with observable criteria: inventory coverage, discovery frequency, owners, runtime enforcement, evidence.
Does an AI workflow need the same governance as an AI agent?
Workflows and agents differ in who picks the next step. A four-question test to classify a system and which agent controls a fixed workflow can drop.
How should you handle exceptions to AI agent policies?
An exception process for AI agents: required fields, compensating controls, approvers by risk tier, expiry of 90 days or less, and risk committee reporting.
Discovery and shadow agents
All in this topicHow do you find AI bots and agents installed in Slack and Microsoft Teams?
Export installed apps from Slack and Teams admin, flag AI apps by the scopes they hold, and use approval and app-centric management to control new installs.
How do you find AI agents running in CI/CD pipelines?
Scan workflow files for AI actions and model secrets, use GitHub allowed-actions policies, and track coding agents triggered from issues and pull requests.
How do you find agentic AI browsers on company devices?
Find AI browsers and browser agents with MDM app inventory, managed-browser extension reports and network logs, then decide to allow, restrict or block each one.
How do you keep your AI agent inventory in sync with the CMDB?
Which CMDB classes to use for AI agents, which fields sync one way vs two way, how to stop discovery scans creating duplicate CIs, and who owns each record.
Agent identity and access
All in this topicShould AI agents use personal access tokens?
Personal access tokens tie an agent to one person's identity and permissions. Compare app installation tokens, bot tokens, workload identity and on-behalf-of access.
How do you revoke all of an AI agent's access at once?
Stop the agent in the request path, then revoke every credential type it holds in the right order, and test that each one now fails.
Should multiple AI agents share one service account?
Shared service accounts widen blast radius, erase attribution and block revoking one agent. How to migrate to one identity per AI agent, step by step.
Can continuous access evaluation revoke an AI agent's access mid-task?
How OpenID Shared Signals and CAEP events like session-revoked reach AI agents, and the gap when long-running tasks hold cached access tokens.
How do you give AI agents credentials without exposing them to the model?
Keep secrets out of prompts: inject credentials at the tool layer, use a token vault and short-lived tokens, and test that no secret ever appears in traces.
How should manual role overrides work when roles come from SCIM?
SCIM keeps roles in sync with your identity provider, but sometimes a role is wrong. Precedence rules, expiry, sync behaviour and audit for manual overrides.
How do you control what an AI agent can retrieve from a vector database?
Vector databases often connect straight to agents, bypassing MCP and gateway controls. How to enforce retrieval permissions server-side and log what was retrieved.
Runtime control
All in this topicWhat is a circuit breaker for AI agents?
A circuit breaker automatically stops an AI agent when error rate, spend or tool failures cross a threshold, then tests recovery. States, triggers and thresholds.
How do you stop an AI agent from emailing people outside the company?
Layer a recipient allowlist in the tool, mailbox-scoped send permissions, a mail flow or delivery rule, and alerts. A system prompt alone does not stop external email.
Can you undo what an AI agent did?
Only some agent actions can be undone. Classify tools as reversible, compensable or irreversible, keep an action journal, and control irreversible tools up front.
How do you test a new AI agent policy before enforcing it?
Run new agent policies in monitor mode on real traffic, measure the false-positive rate over a fixed window, then promote from monitor to warn to block.
How do you reconstruct what an AI agent did during an incident?
The trace data you need to rebuild an AI agent incident, why exact replay rarely works, and how to keep chain of custody for the evidence.
How do you govern an AI agent that extracts data from documents for underwriting?
Controls for underwriting extraction agents: classify first, per-field confidence thresholds, master-data checks and a reviewer approval record that holds up.
Tools and MCP
All in this topicCan you trust MCP tool annotations like readOnlyHint?
MCP tool annotations are hints the spec says to treat as untrusted unless the server is trusted. How to use them in policy without relying on them.
What changed in the July 2026 MCP specification for enterprise security?
MCP 2026-07-28 removes sessions, deprecates Dynamic Client Registration for CIMD, requires iss validation and binds credentials to their issuer.
What is the confused deputy problem in MCP?
How an MCP proxy with a static client ID and a consent cookie can hand an attacker an authorization code, and the per-client consent fix the spec requires.
Gateways and cost
All in this topicHow much do MCP tool definitions add to token costs?
MCP tool schemas are sent as input tokens on every model call. How to measure the overhead per agent, a worked example, and ways to cut it.
Should embedding calls go through the LLM gateway?
Embedding calls send whole documents to a provider, spike during re-indexing and are often missing from cost reports. When to gateway them and when a direct path is fine.
How do you forecast AI agent spend for next year's budget?
A driver-based model for AI agent spend: agents × runs × tokens × price, with scenarios for price drops, model mix and adoption, plus seat lines.
How do reasoning tokens change the cost of an AI agent?
Reasoning models bill hidden thinking tokens as output. How effort and thinking budgets work, how to measure them per task, and a worked cost example.
Why does AI spend in the gateway not match the provider invoice?
Gateway cost figures and model provider invoices rarely match. The usual causes, and a monthly reconciliation checklist with a tolerance threshold.
Evidence and frameworks
All in this topicWhat does the Texas Responsible AI Governance Act mean for companies using AI agents?
TRAIGA (HB 149) took effect 1 January 2026: intent-based prohibitions, AG enforcement with a 60-day cure period, and NIST AI RMF and red-team defences.
What evidence do AI agents need in GxP-regulated life sciences?
21 CFR Part 11, EU GMP Annex 11 and draft Annex 22 mapped to AI agent evidence: versioned registry, audit trails, change control and validation records.
Which ISO/IEC 27001 Annex A controls apply to AI agents?
Map ISO/IEC 27001:2022 Annex A controls such as 5.9, 5.15, 8.2, 8.15 and 8.16 to AI agent evidence auditors can sample, without needing ISO 42001 first.
Buying and comparison
All in this topicHow do you build a business case for AI agent governance?
Five value lines with formulas (unused seats, duplicate tools, model cost, audit hours, incident exposure) and a one-page template for an AI agent governance budget.
What should you ask references when buying an AI agent governance platform?
Fifteen reference-call questions for AI agent governance vendors, grouped by deployment, discovery, enforcement, support and renewal, with red-flag answers.
Do you need an AI agent governance platform if you already have a GRC tool?
GRC tools hold policies, risks and attestations; agent governance platforms produce machine evidence from live systems. How they differ and how to connect them.
Which integrations should an AI agent governance platform have on day one?
A tiered list of day-one integrations for AI agent governance, from identity and clouds to SIEM and GRC, and what each must return to count as covered.
Platform Guides
All in this topicWhat does Microsoft Defender for Cloud AI threat protection detect?
Defender for Cloud AI threat protection has 18 documented alerts for Azure AI workloads. What it detects, prerequisites and the gaps for agents.
What is the Kubernetes Gateway API Inference Extension?
The Gateway API Inference Extension adds model-aware routing to Kubernetes via InferencePool and an Endpoint Picker. What it governs and what not.
Can you enforce Cursor privacy mode for every developer?
Enforce Cursor Privacy Mode in the team dashboard and block personal accounts with Allowed Team IDs via MDM. What it covers, what not, how to verify.
How do you track and cap Salesforce Agentforce consumption costs?
Agentforce Flex Credits bill per action. How Digital Wallet shows consumption by agent, action and channel, and how to set owners, budgets and alerts.
Does the Vertex AI global endpoint break data residency for agents?
Google says the Vertex AI global endpoint gives no data residency guarantee for ML processing. How regional and us/eu endpoints differ and how to pin agents.
Does Azure OpenAI store prompts sent by your agents for abuse monitoring?
How Azure abuse monitoring handles agent prompts, how to apply for modified abuse monitoring, how to verify ContentLogging=false, and what still gets stored.
Which LLM API features are excluded from zero data retention?
Zero data retention covers stateless calls, not stored state. Which OpenAI API endpoints are not ZDR-eligible, and how to keep ZDR agents away from them.
How do you control GitHub Copilot usage-based spend?
Since June 2026 GitHub Copilot bills AI Credits by token use. How pooling, budgets, cost centres and the stop-at-limit setting work, plus a control checklist.
How do you lock Claude Code permissions with managed settings?
Deploy Claude Code managed settings that developers cannot override: file paths per OS, MDM and server-managed delivery, precedence and the key lock settings.
How do you force every Amazon Bedrock model call to use a guardrail?
Use the bedrock:GuardrailIdentifier IAM condition key to make Bedrock guardrails mandatory. Example policy patterns, limits, and the gaps it leaves for agents.
Why can an agent not invoke a Bedrock model you can see in the console?
Bedrock models appear but agent calls fail. The usual causes: Anthropic first-use form, inference profile IDs, destination-Region IAM and SCP Region rules.
Put the framework into practice