---
title: What should an RFP for AI agent governance include?
url: https://opencontroller.lyzr.com/blog/ai-agent-governance-rfp-requirements
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# What should an RFP for AI agent governance include?

**An RFP for AI agent governance should have eight requirement sections: discovery and inventory, identity and access, tools and MCP, runtime control, evaluation, cost, evidence and reporting, and deployment and commercial terms. Mark each requirement "must" or "should", and for every "must" ask for proof you can verify in a demo or pilot, not a yes/no answer.**

Most AI governance RFPs are adapted from data governance or model risk templates. They ask about model cards and bias tests and miss the things that matter for agents: what they can call, whose authority they use, and how you stop one.

## How should the RFP be structured?

Keep it short enough that vendors answer carefully. Around 40 to 60 requirements is usually enough. For each requirement, include:

- **ID and wording.** One testable sentence.
- **Level.** Must or should.
- **Proof requested.** Live demo, pilot test, documentation, or reference architecture.

Ask vendors to answer "available now", "on roadmap with date" or "not planned". A "yes" without a status hides roadmap items.

## What requirements belong in each section?

| Section | Example requirement | Level | Proof to request |
|---|---|---|---|
| Discovery and inventory | Automatically discovers agents, models and API keys in our cloud accounts without installing agents on workloads | Must | Live discovery in a sandbox account during pilot |
| Discovery and inventory | Discovers agents in the third-party SaaS platforms we use | Should | List of supported platforms and what each connector reads |
| Identity and access | Supports delegated on-behalf-of identity so agents act with the user's permissions | Must | Demo of a call made on behalf of a test user, with the audit record |
| Identity and access | Supports SSO via OIDC or SAML, and SCIM provisioning | Must | Configuration guide and test login with our IdP |
| Tools and MCP | Maintains an allow-list of MCP servers and tools per agent | Must | Demo of a blocked, unregistered MCP server |
| Runtime control | Can block or quarantine a single agent without stopping others | Must | Timed demo of quarantine on a test agent |
| Runtime control | Supports policies as code, versioned in our repository | Should | Sample policy and change history |
| Evaluation | Runs multi-turn and red-team evaluations before release | Must | Sample report from a pilot agent |
| Evaluation | Scores a sample of production traffic with LLM judges | Should | Dashboard from pilot traffic |
| Cost | Attributes model cost per agent and per version | Must | Pilot report reconciling to provider invoices |
| Evidence and reporting | Maps controls to the frameworks we report against | Must | Mapping table and exported evidence sample |
| Deployment | Can run in our own cloud account | Must or should | Reference architecture and data flow diagram |

For the evidence section, name your frameworks. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) and [ISO/IEC 42001](https://www.iso.org/standard/81230.html) are common. Ask for "evidence for" each control, not a claim that the product makes you compliant. No tool makes an organisation compliant on its own.

## What commercial questions should the RFP ask?

Agent estates grow fast. Pricing that looks small in year one can scale badly. Ask:

1. **What is the pricing unit?** Seat, agent, request, token or deployment. See [how AI agent governance software is priced](/blog/how-is-ai-agent-governance-software-priced).
2. **What happens to price at three times our current agent count?**
3. **What does implementation include?** Who does connector setup and how long it takes.
4. **What are data residency and hosting options?**
5. **What is the exit plan?** How you export the registry, policies and logs.

## How do you score responses?

Score in two passes. First, any "must" answered "not planned" removes the vendor. Second, weight the remaining scores:

| Area | Suggested weight |
|---|---|
| Discovery and inventory | 15% |
| Identity and access | 15% |
| Runtime control | 20% |
| Evaluation | 10% |
| Tools and MCP | 10% |
| Cost | 10% |
| Evidence and reporting | 10% |
| Deployment and commercial | 10% |

Then verify the top two in a pilot. Written answers rank vendors; pilots decide. Read [how to run a pilot of an AI agent governance platform](/blog/how-to-run-a-pilot-of-an-ai-agent-governance-platform), the [security questions to ask an AI governance vendor](/blog/security-questions-to-ask-an-ai-governance-vendor), and the full [AI agent governance platform evaluation checklist](/blog/ai-agent-governance-platform-evaluation-checklist).

## How Opencontroller does this

Opencontroller, by Lyzr, is built to be tested against an RFP like this. It auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, connects to Salesforce, SAP, Databricks and HubSpot, and supports OIDC SSO and delegated on-behalf-of identity. It keeps a governed registry of agents, tools and MCP servers, runs multi-turn and red-team evals, attributes cost per agent and version, and offers per-agent Block and Quarantine. Policies are code, with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. It runs as Lyzr SaaS or in your own infrastructure, with Lyzr engineers embedded for integration, under an annual licence per deployment. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**How many requirements should an AI governance RFP have?**
Around 40 to 60. More than that and vendors copy marketing text instead of answering precisely.

**Should an RFP ask whether a tool is compliant with the EU AI Act?**
Ask what evidence the tool produces for specific obligations instead. Compliance is an organisational outcome, not a product feature.

**What is the most important RFP requirement for agents?**
The ability to stop or quarantine one agent quickly without disrupting others, demonstrated live.

## Key facts

An AI agent governance RFP should cover eight sections: discovery and inventory, identity and access, tools and MCP, runtime control, evaluation, cost, evidence and reporting, and deployment and commercial terms. Use 40 to 60 testable requirements marked must or should, request verifiable proof (live demo, pilot test, documentation) for each must, and ask for status as available now, roadmap with date, or not planned. Suggested scoring weights: runtime control 20%, discovery 15%, identity 15%, others 10% each. Ask for evidence for frameworks such as NIST AI RMF and ISO/IEC 42001 rather than compliance claims. Opencontroller by Lyzr is licensed annually per deployment, not per seat or usage; pricing starts from $250K per deployment per year.

## Related

- [How should you run a pilot of an AI agent governance platform?](https://opencontroller.lyzr.com/blog/how-to-run-a-pilot-of-an-ai-agent-governance-platform.md)
- [What security questions should you ask an AI agent governance vendor about itself?](https://opencontroller.lyzr.com/blog/security-questions-to-ask-an-ai-governance-vendor.md)
- [How are AI agent governance platforms priced: per seat, per usage or per deployment?](https://opencontroller.lyzr.com/blog/how-is-ai-agent-governance-software-priced.md)
- [More on Buying and comparison](https://opencontroller.lyzr.com/blog/cluster/buying-and-comparison)
