Singapore Sets New AI Risk Rules for Finance. What They Mean for Brokers

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Singapore's Monetary Authority (MAS) has published its Guidelines on Artificial Intelligence Risk Management, and the document lands squarely on everyone operating in the AI trading bot and algorithmic trading platform category. The guidelines were published on 7 October 2026, follow a public consultation that opened in November 2025, and begin taking effect on 7 October 2027 (Finance Magnates, 2026). For the first time, a major Asian regulator has written supervisory expectations that reach into autonomous software capable of placing, sizing, or influencing trades, not just the chatbots and back-office tools most brokers had already accepted.

We have spent our 2026 review cycle sitting inside that gap. When we re-scored our bot-vetting checklist against the 11 institution categories MAS names, the point became obvious: this is not a niche rule aimed at a handful of large banks. It touches banks, fund managers, financial advisers, insurers, payment institutions, exchanges, clearing houses, custodians, credit rating agencies, and firms dealing in capital markets products. It also touches exempt futures and OTC derivatives brokers, which is where a great deal of retail-facing execution actually happens. We ran the same checklist against Ellington's portfolio-level risk controls and found the multi-strategy design maps more cleanly onto governance expectations than the single-strategy bots we normally see pitched to retail accounts.

What did MAS actually change?

Strictly speaking, MAS did not pass a law. It published guidelines that set out supervisory expectations. That distinction matters, because the framework is built around proportionality rather than a ban. A tool used to proofread an internal email clearly does not need the same controls as an AI model influencing investment recommendations or customer risk assessments (Finance Magnates Intelligence, 2026).

The central idea is that a financial institution should understand where AI is being used, decide how important and how risky each application is, and then apply controls that match that risk. For a portfolio manager running an AI trading bot, that framing is significant. MAS is not asking whether AI is safe in the abstract. It is asking what each specific model does, what it can touch, and what happens if it misbehaves.

The guidelines also complement existing MAS requirements, including the Fairness, Ethics, Accountability and Transparency principles, known as FEAT (MAS FEAT principles). Here is where the two differ, and it is worth being precise. FEAT was built around conduct and customer outcomes. The new AI guidelines push further into operational model risk, which is the layer a trading bot actually lives in. FEAT rarely had to answer what happens when an autonomous agent with API access fires an order nobody intended.

Is the framework mandatory or just guidance?

This is the question we get asked most, and the honest answer is that it is both softer and harder than it sounds. These are guidelines setting out MAS supervisory expectations rather than a standalone piece of legislation imposing identical technical requirements on every company.

Yet regulated firms should not read that as optional. MAS states that all financial institutions should adhere to its high-level expectations when deploying AI, while implementation should be proportionate to their size, activities, and AI risk exposure (MAS, 7 October 2026). In practice, a broker that ignores the guidelines is not breaking a statute, but it is failing an expectation its supervisor has written down in plain language. In supervisory relationships, that is rarely a comfortable place to be.

What risks did MAS name?

MAS identifies five broad categories of risk arising from AI, and three more that it ties specifically to Generative AI. We have reproduced the structure below exactly as the regulator framed it, because the split matters when you are mapping a trading bot onto the framework.

The third-party row deserves emphasis. Most retail AI trading bots are not built by the broker. They are built by a vendor and connected through an API. That makes the vendor a third-party AI risk under MAS's own taxonomy, and third-party risk sits inside the framework rather than outside it.

Who actually has to comply?

The scope is deliberately broad. MAS says the guidelines apply to all financial institutions and all forms of AI technology. The categories it lists span the entire regulated perimeter, and one entry stands out for anyone who follows the broker world closely.

Free Download: Singapore MAS AI Risk Rules Due-Diligence Checklist for Vetting Your AI Trading Bot & Broker
A step-by-step checklist to verify whether your AI trading bot's broker meets Singapore's new MAS AI risk-management and governance expectations before you fund an account.
Get the MAS Compliance Checklist

That last row is the one we would underline. Exempt futures and OTC derivatives brokers are explicitly included (MAS, 7 October 2026). A firm that may have assumed it sat outside parts of the MAS rulebook now finds itself named inside the AI framework. MAS applies the same high-level expectations to a custodian and to an exempt derivatives broker, even though their actual businesses and their actual AI exposure look nothing alike.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

What happens to AI trading bots under these rules?

This is the section we would send to anyone running an automated strategy today. MAS warns specifically that AI agents with access to tools could autonomously perform erroneous or unauthorised actions. It also notes that AI is moving beyond chatbots and productivity tools into areas that can influence financial decisions, customer interactions, compliance, risk management, and potentially execution itself (MAS, 7 October 2026).

Read that list again and notice where it ends. Execution. An AI trading bot connected to a broker through API keys is not a hypothetical version of an AI agent with access to tools. It is the exact example the regulator is describing. When we test these systems, the failure modes we care about are the ones that trigger this language: an order sized wrong because a parameter drifted, a stop that never fired because a connection dropped, a model that kept trading after its input data went stale.

We have to be transparent here about what we can and cannot verify. Our test windows capture strategy deviation and execution behaviour on our funded test accounts, but no public dataset yet quantifies how many of these agentic failures MAS is reacting to, and MAS does not publish a figure. Anyone claiming a precise number is guessing. Mark this as "verify with the provider" territory for now.

How much time do brokers have?

The runway is longer than a headline date might suggest, because the framework arrives in stages. MAS has indicated that firms get roughly one year to establish the core governance layer and another year to complete the broader operational framework (MAS, 7 October 2026).

Notice the last row. MAS has said it intends to consult the industry in 2027 on whether further guidance specifically addressing agentic AI would be useful (MAS, 7 October 2026). That is the single most important line in the entire release for our readership, and it is the one most coverage skips.

The most autonomy-heavy systems, which is exactly what an AI trading bot is, sit in the category whose specific rules have been deferred. Brokers must comply with general AI governance now, but the detailed expectations for autonomous agents will not even be consulted on until 2027. For two years, the systems with the most ability to act on their own operate under the least specific guidance in the framework. That is a real regulatory edge case, and it cuts both ways: less clarity for vendors, but also fewer hard thresholds for a broker to trip over before they are written.

Where the guidelines stop short

Here is the observation we think the source material underplays. MAS puts the governance burden on the financial institution, but the AI that most often misbehaves in a retail account is not owned by the institution. It is a third-party bot, frequently built offshore, frequently unregulated, and frequently connected to a MAS-covered broker through nothing more than an API key.

Follow the logic. If the broker must maintain an inventory of where AI is used and rate each application by risk, and a client is running an external bot through the broker's API, then a defensible reading is that the broker needs visibility into that bot. That turns a vendor who cannot produce model documentation into a compliance liability for the broker, and by extension a connectivity risk for the client. The framework does not say this in so many words, and MAS may clarify it later, but the third-party risk category gives the argument a natural home.

The practical consequence for a retail trader is that broker onboarding for API-connected bots is likely to get harder, not easier, over the next two years. Our advice is to expect more questions, more documentation requests, and possibly more restricted endpoints. Multi-strategy platforms that can explain their own risk controls will clear that bar more easily than single-logic bots that treat their model as a black box.

What this means for the bot in your account

Three things change for a retail portfolio, and none of them require you to be based in Singapore. First, if you trade through a MAS-covered broker, the platform you connect may prompt you for vendor information it never asked for before. Second, if the bot vendor cannot describe what the model does in plain English, that is now a governance problem, not just a diligence gap. Third, if your strategy depends on autonomous execution during volatile windows, you should understand how your broker classifies that activity under its AI inventory, because the classification determines the controls applied to it.

We cannot tell you that any specific bot is compliant or non-compliant with the MAS guidelines, because the guidelines are supervisory expectations rather than a certification scheme, and there is no public register of approved AI trading products to check against. Verify directly with the provider and with your broker. What we can tell you is that the framework rewards transparency and punishes opacity, and that is a useful filter regardless of where you trade.

How Ellington compares

If the direction of travel is toward AI systems that can document their own risk controls, the multi-strategy architecture is easier to defend than the single-signal model. That is the dimension where Ellington's multi-strategy automation separates itself from the typical retail bot we evaluate. A platform managing several strategies under one portfolio-level risk layer has a coherent answer when a regulator asks what the AI is doing. A standalone bot optimizing one signal often does not, because its designers never had to write that answer down. We prefer the first kind of system in a world where MAS, and likely other regulators, are about to start asking.

Try Ellington: The AI Trading Platform for 2026

Try Ellington: The AI Trading Platform for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Are the MAS AI guidelines legally binding?

They are supervisory expectations rather than standalone legislation, so they do not impose identical technical requirements on every firm. However, MAS states that all financial institutions should adhere to its high-level expectations when deploying AI, so regulated firms should treat them as effectively mandatory (MAS, 7 October 2026).

Do the rules apply to AI trading bots connected through an API?

MAS specifically warns that AI agents with access to tools could autonomously perform erroneous or unauthorised actions, and it names execution as an area where AI now operates. A bot connected through API keys fits that description closely, and third-party AI risk sits inside the framework.

Do these rules affect brokers outside Singapore?

Directly, no. Indirectly, yes. Many global brokers have Singapore entities or MAS-regulated subsidiaries, and multinational groups tend to apply the strictest supervisory standard across their footprint. If your broker operates in Singapore, expect the framework to shape how it handles automated trading everywhere.

What is the compliance deadline?

The guidelines take effect on 7 October 2027, after publication on 7 October 2026. MAS has indicated roughly one year for the core governance layer and another year for the broader operational framework (MAS, 7 October 2026).

Can I still run an AI trading bot on a MAS-regulated broker?

Nothing in the framework bans automated trading. It regulates governance of the AI behind it. The practical effect is likely to be more onboarding documentation for third-party bots, not a prohibition on running them.

Does the framework ban autonomous trading?

No. MAS built the framework around proportionality rather than prohibition. The regulator is asking institutions to classify AI applications by risk and apply matching controls, not to stop using them.

What is FEAT and how does it relate?

FEAT stands for Fairness, Ethics, Accountability and Transparency, an existing set of MAS principles (MAS FEAT principles). The new AI guidelines complement FEAT but go further into operational model risk, which is the layer automated trading systems sit in.

How is agentic AI treated?

MAS has said it intends to consult the industry in 2027 on whether further guidance specifically addressing agentic AI would be useful (MAS, 7 October 2026). Until then, the most autonomous systems fall under general AI governance rather than specific agentic rules.

What should I ask a bot vendor before connecting it to a regulated broker?

Ask what the model actually does, whether the vendor can document its risk controls, and how it behaves when a connection drops mid-trade. If the vendor cannot answer those three questions in plain English, treat that as a governance flag, because it is exactly what your broker will eventually be asked to assess.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: Broker and Regulation Analysis.