FEAT stands for fairness, ethics, accountability and transparency, the principles MAS set for AI and data analytics in Singapore's financial sector. Applied to AI agents, they become four controls: test outcomes across groups, hold agents to the same ethical standards as staff, name an owner and give customers a route to appeal, and tell people what agents do and why.
What do the FEAT principles require?
The principles are short. In substance:
Fairness. Individuals or groups are not systematically disadvantaged by AI-driven decisions unless justified. Personal attributes used as inputs are justified. Data and models are reviewed for accuracy, relevance and unintended bias, and decisions are reviewed so systems behave as intended.
Ethics. The use of AI is aligned with the firm's ethical standards, values and codes of conduct. AI-driven decisions are held to at least the same ethical standards as decisions made by people.
Accountability. Internally, the use of AI is approved by an appropriate internal authority, and the firm is accountable for models it builds and those it buys; management and the board are kept aware. Externally, customers have channels to enquire about, appeal and request review of AI-driven decisions, and verified, relevant information they supply is taken into account.
Transparency. The use of AI is disclosed proactively as part of general communication, and, on request, customers receive clear explanations of what data is used to make decisions about them, how it affects them, and the consequences.
For the shorter scope question, see how MAS FEAT principles apply to AI agents.
Who do they apply to?
Financial institutions in Singapore that use AI and data analytics, and any firm that adopts them as a reference. MAS expects firms to apply them proportionately: the more material the decision, the more rigour.
What do the principles mean when the system is an agent that acts?
FEAT was written before agents became common, but its wording fits them well, partly because it speaks of "AI-driven decisions" rather than models. Each principle has a sharper edge for agents.
Fairness reaches every step the agent takes. Not only the decision, but who gets escalated, how long they wait and what options they are offered. The method is in MAS FEAT fairness assessment for AI agents.
Ethics means the agent cannot do what a person would not be allowed to do. "Same ethical standards as human-driven decisions" is a practical test. If staff may not pressure a customer in hardship, push a product outside the customer's needs or withhold a cancellation route, neither may the agent. Write those rules into the agent's bounds, not only its instructions, because instructions can be misread and bounds cannot.
Accountability covers bought agents. A firm using a vendor's pre-trained agent is accountable for it as if it had built it. "Approved by an appropriate internal authority" means someone with standing signed off this agent, for these actions, on evidence.
Transparency means the explanation must be about this customer. On request, the customer is owed what data was used about them and how it affected the outcome. For agents, that requires a record of which facts the agent used and where they came from.
How do you turn FEAT into controls?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Fairness: no unjustified disadvantage | Outcome testing by group on real cases; live monitoring | Test and monitoring reports |
| Fairness: justified attributes | Register of personal attributes the agent can see, with reasons | Attribute register |
| Ethics: same standards as people | Conduct rules for staff mapped into agent bounds; prohibited behaviours tested | Bounds configuration; test results |
| Ethics: alignment with values | Agent use reviewed by an ethics or conduct forum for material decisions | Forum minutes |
| Accountability: internal approval | Named owner and approving authority per agent capability | Approval records |
| Accountability: bought models | Vendor agents held to the same approval and testing as in-house | Due diligence; test results |
| Accountability: board awareness | Agent use and incidents reported to management and board | Board papers |
| Accountability: customer appeal | Route to a person for enquiry, appeal and review; new information considered | Appeal log with outcomes |
| Transparency: proactive disclosure | Customers told where agents are used | Disclosures |
| Transparency: explanation on request | Per-decision record of facts used and their source | Decision records |
How does FEAT compare with other principle sets?
FEAT has much in common with data protection principles, such as GDPR's Article 5 requirements for fairness, transparency and accuracy; see GDPR Article 5 principles applied to AI agents. The difference is that FEAT is specific to financial services decisions and adds the ethics test. Both leave the design of controls to the firm, which is the subject of our argument that principles-based regulation leaves the agent design to you.
What will MAS ask?
- Which agents make or shape decisions about customers?
- Who approved each, and on what evidence?
- How do you ensure the agent meets the same conduct standards as staff?
- What channel does a customer use to appeal, and what happens next?
- Can you explain to a customer what data drove a decision about them?
- Is the board informed about agent use and incidents?
What do the principles not cover?
They are not binding rules with penalties, though MAS expects them to be applied. They do not replace the Technology Risk Management Guidelines, the PDPA or conduct rules. They do not prescribe metrics or tests. For the full map, see AI agent regulation, mapped to controls.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Questions people ask
Does FEAT require a person to review every agent decision?
No. It requires accountability, a route to appeal and review, and decisions held to the same standards as people's. The level of human involvement should match materiality.
Who should be the "appropriate internal authority"?
Someone with authority over the business process and the risk, such as a business head with risk sign-off. For material decisions, many firms add a committee.
Is FEAT relevant outside Singapore?
As a reference, yes. Its principles are clear and practical, and they align with expectations in other markets.