---
title: What evidence do the MAS AI risk management guidelines expect for AI agents?
url: https://opencontroller.lyzr.com/blog/mas-ai-risk-management-guidelines-evidence-for-agents
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# What evidence do the MAS AI risk management guidelines expect for AI agents?

**The Monetary Authority of Singapore (MAS) published its Guidelines on Artificial Intelligence Risk Management on 7 October 2026. They take effect on 7 October 2027 and apply to all financial institutions and all forms of AI, including agents. For AI agents, MAS expects five kinds of evidence: board oversight, an AI inventory, a risk materiality assessment per use case, proportionate lifecycle controls, and accountability for third-party AI.**

MAS also noted the growing use of agentic AI systems that act autonomously and use tools. It plans to consult the industry in 2027 on whether more agent-specific guidance is needed. Until then, these guidelines are the baseline.

## What are the key dates?

| Date | What happens |
|---|---|
| November 2025 | Consultation paper on proposed guidelines |
| 7 October 2026 | Final guidelines published |
| 7 October 2027 | Guidelines take effect; expectations in Sections 3 and 4 apply |
| 7 October 2028 | Expectations in Sections 5 and 6 due |
| 2027 (planned) | MAS consultation on additional agentic AI guidance |

The scope covers banks, finance companies, capital markets intermediaries, payment institutions, insurers and other regulated entities. Check the guidelines for which expectations sit in each section.

## What does MAS expect, and what does that mean for agents?

**1. Board and senior management oversight.** Boards and senior management should oversee AI risk: roles, risk appetite and the risk management framework. MAS says existing governance structures can be used if they give adequate oversight. A new AI committee is not required.

For agents, the board needs to see how many agents act on the firm's behalf, which ones are material, and what happens when one misbehaves.

**2. An AI inventory.** Firms should identify their AI use and keep an inventory at a suitable level of detail. Agents are hard to inventory. They appear inside SaaS platforms, cloud projects and developer tools, and they change version often. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).

**3. Risk materiality assessment.** Each AI use case should get a materiality assessment. The consultation framed materiality around impact, complexity and the degree of reliance on the AI. An agent that moves money or changes customer records is more material than one that drafts internal notes.

**4. Lifecycle controls.** Controls should be proportionate and cover data governance, testing, human oversight, cybersecurity, monitoring and change management. Firms should review them as AI use grows. Low-impact uses may need only basic policies and procedures.

**5. Third-party AI.** Firms stay accountable for AI they use, including AI from vendors. They should get enough assurance, assess suitability for the intended use and add compensating controls where there are gaps. If risk falls outside appetite, they should consider limiting, suspending or replacing the service.

## MAS evidence checklist for AI agents

Use this to prepare before 7 October 2027.

1. **Inventory**: every agent, its platform, owner, version, model, tools and data sources. Include agents in third-party SaaS.
2. **Materiality rating**: per agent use case, with the reasoning recorded and a review date.
3. **Board reporting**: a regular summary of agent count, material agents, incidents and exceptions.
4. **Testing records**: pre-deployment and periodic evaluations, including adversarial (red-team) testing for material agents. See [how to red-team AI agents](/blog/how-to-red-team-ai-agents-multi-turn).
5. **Human oversight**: who reviews which agent actions, and the record of reviews. See [human review evidence](/blog/evidence-for-human-review-of-ai-agent-outputs).
6. **Cybersecurity**: agent identities, permissions, secrets and tool access, with risky configurations flagged.
7. **Monitoring**: production behaviour, drift and alerts, with the ability to suspend an agent quickly.
8. **Change management**: version history showing what changed (prompt, model, tools) and who approved it.
9. **Third-party assurance**: vendor assessments for agent platforms and models, and the compensating controls you added.
10. **Exit and suspension plan**: how you would limit or switch off a third-party agent service.

## How is this different from Singapore's other agentic AI guidance?

The MAS guidelines are supervisory expectations for regulated financial institutions. IMDA's Model AI Governance Framework for agentic AI is voluntary guidance for any organisation. Firms regulated by MAS will usually use both. See [Singapore's agentic AI governance guidance](/blog/singapore-agentic-ai-governance-guidance).

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, and connects to Salesforce, SAP, Databricks and HubSpot, plus Claude Enterprise and OpenAI Enterprise workspaces. Every agent, version, tool and MCP server goes into a governed registry. Posture flags risky permissions and configurations. Multi-turn and red-team evals, and LLM judges on production traffic, produce testing evidence. Rogue-agent handling monitors for behavioural deviation and can quarantine or kill an agent; per-agent Block and Quarantine act through the connected platform. These outputs are **evidence for** the MAS expectations, not a substitute for the firm's own risk judgement. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

Read next: [AI agent controls for banks](/blog/ai-agent-controls-for-banks-model-risk) and [how to map agent controls to frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## Frequently asked questions

**When do the MAS AI risk management guidelines take effect?**
On 7 October 2027. MAS published them on 7 October 2026 and phased the rollout, with some sections due by 7 October 2028.

**Do the MAS guidelines cover AI agents?**
Yes. They apply to all forms of AI. MAS noted the growing use of agentic AI and plans to consult in 2027 on whether more agent-specific guidance is needed.

**Does a financial institution need a separate AI committee under the MAS guidelines?**
No. MAS says existing governance structures can be used if they provide adequate oversight of AI risk.

## Key facts

MAS Guidelines on Artificial Intelligence Risk Management: consultation Nov 2025; final published 7 October 2026; effective 7 October 2027 (Sections 3-4), Sections 5-6 by 7 October 2028; apply to all financial institutions and all forms of AI. Expectations: board oversight (existing structures allowed), AI inventory, risk materiality per use case, proportionate lifecycle controls (data, testing, human oversight, cybersecurity, monitoring, change management), third-party accountability. MAS plans a 2027 consultation on agentic AI guidance. Opencontroller discovery, registry, Posture, evals and rogue-agent handling provide evidence for these expectations.

## Related

- [What does Singapore's agentic AI governance guidance expect from enterprises?](https://opencontroller.lyzr.com/blog/singapore-agentic-ai-governance-guidance.md)
- [What AI agent controls do banks need now that model risk guidance excludes agentic AI?](https://opencontroller.lyzr.com/blog/ai-agent-controls-for-banks-model-risk.md)
- [What evidence will a regulator ask for about your AI agents?](https://opencontroller.lyzr.com/blog/regulator-inspection-evidence-pack-for-ai-agents.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
