Most AI ethics committees were created in response to a specific moment of pressure: an investor asked a due diligence question about AI ethics, a regulator mentioned oversight requirements, or a competitor announced an ethics board and the leadership team felt it should have one too.
The result, in most cases, is a committee with a vague mandate, a terms of reference document that nobody remembers writing, no scheduled meetings, no defined inputs or outputs, and members who were selected for their seniority rather than for any specific knowledge of AI governance. When a real governance question arises, the committee has no process to follow and no authority to act.
This article explains what separates an ethics committee that provides genuine oversight from one that exists only on paper.
The three signs that an ethics committee is not functioning
It has never produced a written output that affected a decision. A functioning ethics committee reviews AI deployments, assesses risks, and makes recommendations that are acted on. If your committee has met but never produced a document that changed what the organisation did, it is not providing governance. It is providing cover.
Its members do not know what they are supposed to be reviewing. Ask your ethics committee members what they are responsible for overseeing. If they give different answers, or vague answers, or cannot name specific AI systems that fall within their remit, the committee does not have a functional mandate.
There is no reporting line to the board. MAS AIRG is explicit that AI risk must be managed at the board level. An ethics committee that reports to a technical team, or that produces a summary for inclusion in an annual report, is not providing the board-level oversight that regulators expect.
What a functioning ethics committee requires
A specific mandate
The mandate should answer four questions: what AI systems or decisions is the committee responsible for overseeing, what authority does it have to require changes, who does it report to, and what happens when the committee’s recommendation is not followed.
Without answers to all four questions, the mandate is incomplete. A committee with a vague aspiration to “promote responsible AI” has no way to know whether it is succeeding.
The right members
The composition question is usually approached from the wrong direction. Organisations ask: who is senior enough to sit on an AI ethics committee? The right question is: what knowledge and perspective is needed to review AI risk in our context, and who has it?
A functioning ethics committee for a financial services company needs someone with risk management expertise. It needs someone with customer or community perspective. It needs someone with technical AI knowledge, but not necessarily the most senior technical person in the company. And it needs at least one genuinely independent external member who can raise concerns that insiders might not.
External members are particularly important and particularly underused. An external member who brings genuine independence and relevant expertise changes the dynamic of an ethics committee entirely. They are not subject to the internal social pressures that make insiders reluctant to flag problems.
A structured meeting process
Ethics committees that have no agenda template, no defined inputs, and no defined outputs do not produce useful governance. What does a useful meeting look like?
Before the meeting: a summary of any new AI deployments since the last meeting, any incidents or near-misses, any regulatory developments that affect the organisation’s AI use, and any items escalated from operational teams.
During the meeting: a structured review of each item, a clear decision or recommendation for each, and a record of dissenting views if any members disagree with the majority recommendation.
After the meeting: a written summary of decisions and recommendations, circulated to the board or relevant governance body, with a follow-up mechanism to check that recommended actions are taken.
Genuine board-level reporting
The committee’s output needs to reach the board in a form that enables the board to make decisions. A two-paragraph summary buried in a board pack does not achieve this. A structured report covering the AI risk landscape, the committee’s recent recommendations, the organisation’s response to those recommendations, and any open governance concerns gives the board the information it needs to exercise its oversight role.
The workforce transition dimension
Singapore’s AI transformation is changing how work is done across the economy. AI is automating tasks, changing job roles, and creating new accountability questions around decisions that were previously made by humans.
Most AI ethics committees were not designed to review workforce AI impacts. They were designed to review product and customer-facing AI. But the governance question around AI-driven workforce decisions is a legitimate ethics question: who is accountable when an AI system influences a hiring decision, a performance assessment, or a redundancy process?
Extending an ethics committee’s mandate to cover workforce AI decisions, and ensuring the committee has the perspective needed to evaluate those decisions fairly, is a governance gap that very few organisations have addressed. It is one that boards and leadership teams are increasingly aware of, because they are being asked about it by employees, shareholders, and in some cases regulators.
Getting started
If your organisation does not have an AI ethics committee and should have one, the right starting point is designing the mandate before you recruit the members. Too many committees are composed of available and willing people who are then handed a vague terms of reference and expected to figure out what they are supposed to do.
If your organisation has an ethics committee that is not functioning effectively, a governance audit of the committee itself is often more useful than trying to revive it through incremental changes. Understanding specifically why it is not functioning — whether it is a mandate problem, a composition problem, a process problem, or a reporting problem — determines what the right fix is.
Either way, the question to ask is: if our ethics committee had to demonstrate genuine oversight to a regulator tomorrow, what would we show them? The honest answer to that question is usually a better guide to what needs to change than any generic framework document.
Aivance’s Ethics Board Advisory service designs and implements AI ethics committees from mandate through inaugural meeting. If you are building or rebuilding ethics oversight, the AI Governance Review is the right first step.