---
title: What should regulated firms require from an AI agent governance platform?
url: https://opencontroller.lyzr.com/blog/ai-governance-platform-requirements-for-regulated-industries
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# What should regulated firms require from an AI agent governance platform?

**Banks, insurers and healthcare firms need more than features. They need evidence they can hand to a supervisor, control over where governance data lives and for how long, segregation of duties inside the tool, contractual audit and access rights, and a tested way out. Treat the governance platform as an ICT service that may support a critical function, and buy it that way.**

A governance platform for AI agents sees prompts, tool calls, identities and policy decisions. In a regulated firm, that makes it both a control and a source of risk. The general [evaluation checklist](/blog/ai-agent-governance-platform-evaluation-checklist) covers discovery, runtime control and cost. This page adds the requirements that come from supervisors, not from engineers.

## Why are the requirements different in regulated industries?

Three things change. First, supervisors ask for evidence, not intent. You must show an inventory, the policy that applied and the decision that was made, often months later. Second, outsourcing and third-party rules apply to the vendor. In the EU, the Digital Operational Resilience Act (DORA, Regulation (EU) 2022/2554) has applied since 17 January 2025. Its Articles 28 to 30 set out how financial entities manage ICT third-party risk, and Article 30 lists contract terms such as audit rights and exit strategies for services that support critical or important functions. Third, sector data rules travel with the data. A US healthcare firm that lets a vendor handle protected health information needs a business associate agreement under HIPAA.

None of this makes a tool "compliant". The platform produces evidence for your controls. Your compliance team still owns the mapping.

## What is the checklist?

Use these twelve criteria in your RFP and proof of concept. Each one has a concrete test.

1. **Evidence export.** Can you export inventory, policy versions, decisions and evaluation results as files or through an API, with timestamps and the identity of the actor? Test: ask for the record of one blocked action from 30 days ago.
2. **Framework mappings you can edit.** Built-in mappings help, but supervisors use their own taxonomies. Test: add a custom control ID and map three policies to it.
3. **Data residency.** Can the platform and its logs stay in a region you choose, or run on your infrastructure? Test: list every sub-processor and region that touches prompt data.
4. **Retention you set.** Different records need different periods. Test: configure a shorter period for prompt content and a longer one for decision logs.
5. **Redaction before storage.** Test: send a synthetic account number through the path and confirm what is stored.
6. **Segregation of duties.** The person who writes a policy should not approve and deploy it alone. Test: try to publish a policy change with a single admin account.
7. **Immutable admin audit trail.** Every change to policies, connectors and roles should be logged. Test: change a connector credential and find the log entry.
8. **Supervisory and internal audit access.** Test: confirm the contract gives your auditors and your supervisor access to relevant information and premises where the law requires it.
9. **Incident notification terms.** Test: check how fast the vendor must tell you about an incident affecting your data.
10. **Kill and quarantine you control.** You must be able to stop an agent without a vendor ticket. Test: quarantine a test agent and time it.
11. **Exit plan.** Test: export the full registry and policies in an open format and load them elsewhere.
12. **Concentration view.** Test: can the platform show which agents depend on which model provider, so you can assess concentration risk?

## How do requirements differ by sector?

| Sector | Typical source of the requirement | What it means for the platform |
|---|---|---|
| EU financial services | DORA Articles 28 to 30; EU AI Act for high-risk uses | Register of ICT arrangements, contract terms, exit strategy, evidence of oversight |
| Banking (model risk) | Supervisory model risk guidance | Inventory, validation records, ongoing monitoring of agent behaviour |
| Insurance | Insurance supervisors' AI and outsourcing expectations | Fairness and outcome monitoring evidence, vendor oversight |
| US healthcare | HIPAA business associate rules | BAA, minimum-necessary data handling, access logging |
| Life sciences | GxP data integrity expectations | Validated configuration, audit trail, change control |

For sector detail, see [AI agent controls for banks](/blog/ai-agent-controls-for-banks-model-risk), [evidence for insurers](/blog/ai-governance-evidence-for-insurers) and [GxP evidence](/blog/gxp-evidence-for-ai-agents-in-life-sciences).

## SaaS or self-hosted for a regulated firm?

Both can work. Self-hosting keeps prompt data and logs inside your perimeter and simplifies supervisory access. Managed SaaS can still pass if residency, audit rights and exit support are in the contract. Decide by the criticality of the functions your agents support, not by habit. The trade-offs are covered in [SaaS vs self-hosted governance platforms](/blog/saas-vs-self-hosted-ai-governance-platform).

## How Opencontroller does this

Opencontroller runs as Lyzr SaaS or on customer-controlled infrastructure, and Lyzr engineers embed for integration. Policies are code, extensible, with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001 that you can extend. Per-agent Block and Quarantine controls act through the connected platform, and rogue-agent handling can quarantine and kill automatically. It uses open standards (OIDC/SAML/SCIM, OpenTelemetry, REST and webhooks), which helps with export and exit. Licensing is an annual licence per deployment, not per seat or usage. To see it against your own checklist, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does a governance platform make us compliant with DORA or the EU AI Act?**
No. It produces evidence for your controls, such as inventories, policy decisions and logs. Your compliance team maps that evidence to the rules that apply.

**Is the governance platform itself an ICT third-party service under DORA?**
Usually yes for EU financial entities, because it is an ICT service. Whether it supports a critical or important function depends on how you use it, so assess it in your register.

**What is the single most useful test in a pilot?**
Ask for the full record of one blocked action from weeks earlier: the agent, the identity, the policy version and the decision. If that takes days, the evidence model is weak.

## Key facts

Regulated firms buying an AI agent governance platform should require: exportable evidence mapped to the frameworks their supervisors use; data residency and deployment choice; configurable retention; segregation of duties inside the tool; contractual audit and access rights; and a tested exit plan. DORA (Regulation (EU) 2022/2554) has applied since 17 January 2025, and its Article 30 lists contract terms for ICT services supporting critical or important functions, including audit rights and exit strategies. Opencontroller by Lyzr deploys as Lyzr SaaS or on customer-controlled infrastructure, has policy-as-code with mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, and is licensed per deployment per year. Pricing starts from $250K per deployment per year.

## Related

- [How to evaluate an AI agent governance platform: 30 questions for your RFP (2026)](https://opencontroller.lyzr.com/blog/ai-agent-governance-platform-evaluation-checklist.md)
- [How does DORA apply to AI agents in financial services?](https://opencontroller.lyzr.com/blog/dora-and-ai-agents-in-financial-services.md)
- [Should you run an AI governance platform as SaaS or self-hosted?](https://opencontroller.lyzr.com/blog/saas-vs-self-hosted-ai-governance-platform.md)
- [More on Buying and comparison](https://opencontroller.lyzr.com/blog/cluster/buying-and-comparison)
