Opencontroller by Lyzr
Opencontroller blog
Answers to the questions enterprises ask about governing AI agents
Agent governance basics
All in this topicDo read-only AI agents need governance?
Yes, but a lighter set. Read-only agents can still leak data through output, overshare, fall to the lethal trifecta and run up cost. Risks and a control table.
Are system prompt instructions a security control for AI agents?
No. System prompts guide behaviour but can be bypassed by prompt injection. Compare instructions, guardrails, permissions and network controls, and where each fits.
What is agent washing, and how do you spot it?
Agent washing is rebranding chatbots, assistants or RPA as AI agents. Test questions on autonomy, tools and memory, and why it matters for inventory and risk.
Discovery and shadow agents
All in this topicWhat should you do in the first 30 days of building an AI agent inventory?
A week-by-week 30-day plan for an AI agent inventory: cloud accounts, SaaS agent platforms, identity logs and developer devices, with exit criteria per week.
How do you find LLM calls hidden in Airflow and dbt pipelines?
Detection patterns for Airflow (common.ai provider, @task.llm, @task.agent, LLM connections) and dbt (AI SQL functions, Python models), plus owner and identity checks.
How do you find AI SQL functions running in your data warehouse?
Function names to search for in Snowflake, Databricks and BigQuery, the query-history tables to search, and how to turn findings into owned, governed use.
How do you find AI API calls hidden in Apps Script and Office macros?
Spreadsheets can call LLM APIs with personal keys. Detection patterns for Apps Script UrlFetchApp, VBA HTTP objects and Office Scripts fetch, plus remediation.
How do you tell when external AI agents are calling your APIs and website?
Identify outside AI crawlers and agents by user-agent tokens, Web Bot Auth signatures (RFC 9421), IP ranges and behaviour, then set a policy.
How do you find AI agents that contractors use on your data?
Signals that reveal contractor AI use: guest accounts, OAuth grants, unmanaged devices, repo tokens. Plus contract clauses and an onboarding checklist.
How do you inventory Agentforce agents across many Salesforce orgs?
Which metadata types define an Agentforce agent (AiAgentDefinition from API 68.0), how to retrieve them per org, and how to merge orgs into one registry.
How do you find OpenAI Assistants and vector stores left after the Assistants API sunset?
The Assistants API shut down on 26 August 2026. How to find leftover vector stores, files and broken code per project, and clean them up.
Agent identity and access
All in this topicHow long should an AI agent access token live?
Practical lifetimes for AI agent access and refresh tokens by risk tier, with platform defaults, RFC 9700 guidance and how revocation latency changes the answer.
Can AI agents authenticate with SAML?
SAML is built for browser sign-in, so agents need OAuth tokens. How the SAML 2.0 bearer assertion grant (RFC 7522) and token exchange (RFC 8693) bridge the gap.
How can AI agents call Amazon Bedrock from outside AWS without long-term keys?
Credential options for agents and gateways calling Amazon Bedrock from outside AWS: IAM Roles Anywhere, web identity federation and short-term Bedrock API keys.
Runtime control
All in this topicHow do you measure and tune guardrail false positives?
A step-by-step loop for measuring AI guardrail false positives: labelled samples, precision and recall per policy, monitor mode, thresholds and appeals.
Should AI agent guardrails be rules or models?
Deterministic rules, small classifiers and LLM judges compared on latency, cost, explainability and false positives, plus a layered guardrail pattern.
What timeouts should you set for AI agent tool calls?
A timeout policy for AI agent tool calls: per-tool-class limits, total task budgets, partial results, retries with backoff and MCP cancellation.
How do you detect an AI agent trying to expand its own permissions?
Signals that an AI agent is escalating its own access: IAM API calls, OAuth scope requests, config edits, new credentials. Detection rules and responses.
How do idempotency keys stop AI agents repeating actions?
Retries at the model, agent and gateway layers can make AI agents create duplicates. How idempotency keys and idempotent tool design prevent it.
What is dry-run mode for AI agents, and when should you use it?
Dry-run mode lets an AI agent plan real actions without executing them. How it differs from shadow mode, how to build it and when to exit it.
Tools and MCP
All in this topicShould MCP access be authorised per server or per tool?
Server-level MCP access decides who reaches a server; tool-level authorisation decides which actions they can take. What each stops, and interim controls.
Should MCP servers run as sidecars or shared services on Kubernetes?
Sidecar vs shared-service MCP servers on Kubernetes: how each affects isolation, identity, scaling and audit, and when to pick which.
Is it safe to auto-generate MCP servers from OpenAPI specs?
Auto-converting OpenAPI specs to MCP servers risks tool explosion, exposed destructive endpoints, weak descriptions and auth passthrough. A curation checklist.
What are MCP output schemas, and why do they matter for governance?
MCP output schemas let tools declare the shape of their results. What the spec requires, how clients validate structuredContent, and governance uses.
How do you govern browser automation MCP servers like Playwright MCP?
Playwright MCP gives agents a real browser. Settings to restrict origins, isolate profiles, hide secrets and log sessions, plus a policy for developer use.
Gateways and cost
All in this topicHow much does storing AI agent traces cost?
A formula and worked example for AI agent trace volume, why payloads dominate size, and how sampling and tiering cut cost without losing audit evidence.
How do you attribute GPU costs of self-hosted models to AI agents?
Split self-hosted GPU cost across AI agents using weighted token shares from gateway logs, with a worked example and three ways to handle idle capacity.
Can an LLM gateway govern realtime voice APIs?
Realtime voice agents use WebRTC, WebSocket and SIP, not plain HTTP. What a gateway can see and enforce for each transport, and the control points that remain.
How are images and documents priced in AI agent calls?
Images and PDFs are billed as input tokens. The Claude formula, worked examples for screenshot agents, and the levers that cut multimodal agent cost.
Do structured outputs work the same across LLM providers behind a gateway?
OpenAI, Anthropic and Gemini each support a different JSON Schema subset for structured outputs. What breaks when a gateway fails over, and how to plan for it.
Evidence and frameworks
All in this topicWhat does internal audit look for in AI agent governance?
A checklist of what internal audit tests in an AI agent governance review, by control area, with the test steps and the evidence to have ready.
What does NIST AI 100-2 say about attacks on AI agents?
NIST AI 100-2 E2025 classifies adversarial attacks on AI, adds an AI agent security section and expands indirect prompt injection. How to use it for agent tests.
What does the CISA AI Data Security guidance mean for AI agents?
The May 2025 joint CISA, NSA and FBI AI Data Security information sheet, mapped to AI agent data flows: RAG, memory, tool outputs and logs, with evidence to keep.
Buying and comparison
All in this topicWhat data does an AI governance platform store about your agents?
Metadata, traces, prompts and keys: what an AI governance platform stores, where, for how long, and the questions to ask in a data-flow review.
How do you test the scalability of an AI agent governance platform?
A test plan for AI agent governance platform scale: inventory size, discovery run time, event ingest, gateway throughput and added p95 latency, with pass criteria.
What should you ask about an AI governance platform's connectors?
Connectors differ in depth: register, observe or enforce. Questions to ask, acceptable answers and demo tests for AI agent governance platform connectors.
What should regulated firms require from an AI agent governance platform?
A checklist for banks, insurers and healthcare firms: evidence export, data residency, retention, segregation of duties, audit rights and exit terms.
Should you run an AI governance platform as SaaS or self-hosted?
SaaS or self-hosted AI agent governance? A decision table on data sensitivity, network paths, ops effort, upgrades and regulators, plus vendor questions.
Platform Guides
All in this topicHow do you govern agents built in Joule Studio on SAP Build?
Controls for custom Joule Studio agents: who can build, which tools they reach, grounding data, identity propagation, release and tracing on SAP AI Core.
How do you govern Salesforce hosted MCP servers?
Salesforce hosted MCP servers (GA April 2026): External Client App, mcp_api scope, user permissions, enabling servers in Setup, and an 8-step baseline.
How do you govern the Atlassian Remote MCP Server for Jira and Confluence?
Atlassian MCP server admin controls: OAuth 2.1 domain allowlist, API token toggle, IP allowlists, user-scoped permissions, and a 7-step policy baseline.
How do you govern Snowflake Intelligence agents?
Snowflake Intelligence agents use the user's credentials. The grants that control access, the defaults to change, and Restricted Session Scopes.
How do you govern agents built with Databricks Agent Bricks?
Agent Bricks governance: agent types, Can Manage vs Can Query, per-subagent grants, Unity Catalog and Unity Gateway, MLflow evaluation, and a checklist.
How do ServiceNow AI agents connect to external agents via MCP and A2A?
ServiceNow MCP Server, MCP Client and A2A explained, with the controls for each direction: tool exposure, server approval, credentials and audit.
What does ServiceNow AI Control Tower cover for AI agent governance?
ServiceNow AI Control Tower: CMDB-based AI inventory, connectors, lifecycle and risk, AI Gateway for MCP. What it depends on and a 7-step PoC checklist.
How does inbound authentication work for Amazon Bedrock AgentCore Runtime agents?
AgentCore Runtime takes IAM SigV4 or a JWT authorizer, never both. What the JWT checks, how user identity flows to tools, and a review checklist.
How do you govern Cursor background agents?
Cursor Cloud Agents auto-run commands in cloud VMs. A 12-point checklist: Git scope, egress lock, Runtime Secrets, Privacy Mode, review and retention.
How do you track Claude Code usage across a team with the Analytics API?
What the Claude Code Analytics Admin API returns per user per day, what it misses (subscriptions, Bedrock, Vertex), and how to build a governance report.
What does the OpenAI Responses API store, and how do you turn it off for agents?
What the OpenAI Responses API retains by default (30-day application state), how store=false and encrypted reasoning work, and what ZDR changes for agents.
How do you add guardrails to LangChain agents with middleware?
Which LangChain built-in middleware act as guardrails (PII, human-in-the-loop, call limits, fallback), how hooks work, and what still needs central control.
Put the framework into practice