The State of AI Answer Accuracy in Malaysian Insurance, A Preliminary Barometer

Ask a public AI assistant how long you have to wait before escalating an unresolved insurance complaint, which office inside your insurer you are supposed to write to first, or whether the protection scheme would cover the death benefit sitting inside your investment-linked policy, and the answer comes back fast and confident, often with a tidy number or a named step. The fact it is describing is usually a public one: the regulator's own published complaint process, or the protection scheme's own list of what it covers. As of August 2026, this is our third monthly reading, and we again checked answers against those published facts, one by one.

The certified findings in this reading concern the steps a consumer takes when something has gone wrong: how long to wait before escalating, which office to write to first, and what the protection scheme pays. Accurate handling was also present in the reviewed examples. This is one of several sector readings in our barometer, and it sits alongside our Singapore insurance reading as a companion view of the same problem.

How we checked AI answers about Malaysian insurance

In August 2026 we again put high-intent questions, the kind a person types before complaining, escalating, or working out what their protection covers, to a set of public AI systems, among them ChatGPT, Gemini, Copilot, Perplexity, Google AI Mode and Google AI Overview. Questions about how to complain and how long to wait before escalating were compared against Bank Negara Malaysia's own published complaint process. Questions about what the public protection scheme covers were checked against the scheme's own official page: PIDM and its Takaful and Insurance Benefits Protection System (TIPS). Where an answer conflicted with the published record, we flagged it for review, and we hand-verified each finding below against a dated official-source capture before standing behind it.

This is a preliminary reading, and we would rather say so plainly than dress it up. It covers a single month, so it is a directional reading rather than a trend, and the examples we reviewed were selected for diversity, not drawn as a random or complete sample, so we make no claim about how often any error occurs. We report findings, not a scoreboard. We name the AI systems only as the set we tested, never to rank one against another. The public schemes and regulators, the protection scheme and its administrator, the national regulator and its contact centre, and the financial ombudsman, are industry-wide public bodies, so we name them and cite their official pages directly. The insurers and takaful operators we do not name, to preserve the anonymised research design. We retain the underlying evidence and result identifiers internally for audit and right-to-reply, and where a finding turned on a single answer we say so.

Where AI answers about Malaysian insurance drift

The certified findings in this reading each turn on a published fact restated wrongly, or a consumer pointed at the wrong step. In each case the record existed: Bank Negara Malaysia had published the complaint process, and the protection scheme had published what it covers.

On escalation timing. A person asked how long they had to wait for a response before taking an unresolved complaint further. A public AI assistant answered that the general rule is to wait 60 calendar days before escalating, and repeated that the key waiting period is 60 calendar days. Bank Negara Malaysia's published complaint process permits referral to BNMLINK, the regulator's public contact centre, after 14 days. The direction of the error is the one that delays a consumer: an answer that sets the wait at 60 calendar days, on a matter the consumer is already unhappy about, defers a step the published process makes available much sooner.

On where a complaint begins. A person described having emailed their insurer's Claims Unit, and asked whether they were on the right track. A public AI assistant told them that emailing the Claims Unit was the correct first step. The regulator's published process states the opposite on this exact point: the Business Unit or Claims Unit is not the Complaints Unit, and a complaint must first be referred to the insurer's designated Complaints Unit. Writing to the Claims Unit may not satisfy the prerequisite for subsequent referral to BNMLINK, so a consumer who believes they have begun the process may not have.

On the healthcare protection limit. A person asked what the public protection scheme covers. A public AI assistant said healthcare benefits were capped at RM500,000 per individual, and in the same answer quoted the scheme's own protection table, which lists healthcare at 100 percent of the amount payable. Healthcare sits on its own line in that table, with a different limit from the RM500,000 that applies to death and related benefits. The answer imported the death-benefit figure onto the healthcare line and contradicted the source it had just cited. Understating protection has its own cost: a policyholder who believes their healthcare cover is capped where it is not may act on a worse understanding of their position than the scheme gives them.

On the unit-portion death benefit. A person asked whether the death benefit inside an investment-linked policy is covered by the scheme. A public AI assistant said any portion of the payout tied to the market value of the investment units is not protected, and that the death benefit is protected only as the guaranteed sum assured. The scheme's own page draws the line differently. Maturity, surrender and income benefits payable from the unit portion of an investment-linked policy are excluded, but misfortune benefits, including death benefits, payable from that same unit portion are protected under TIPS. The distinction the scheme draws is which benefit, not which portion. This is a technical boundary a policyholder cannot easily check for themselves. We record this as a single answer, and we do not imply it appeared across systems.

Even-handed note, accurate handling was also present in the reviewed examples. A review is only as trustworthy as the answers it confirms and the flags it declines to run, so this belongs alongside the findings above. On the public ombudsman's third-party scope, a reviewed answer correctly distinguished property-damage disputes, which the ombudsman can mediate, from bodily-injury and death matters, which are directed to the courts. On a published policy free-look window, a reviewed answer gave the period correctly with a defensible account of when it starts. And on the external escalation body, a reviewed answer named it correctly and was current on the rebrand from the former Ombudsman for Financial Services to the Financial Markets Ombudsman Service (FMOS). Flags that did not hold on checking against the published source were cleared, including honest non-answers where an assistant declined to guess rather than invent a figure.

Why AI answer accuracy is a governance risk for insurers

None of these answers were written by the insurers, the regulator or the scheme, and none of them can be edited the way an institution's own website copy can. That is the difficulty. Bank Negara Malaysia can keep its complaint process exact, the protection scheme can keep its coverage table clear, and an insurer can keep its own complaints page accurate, and a consumer can still arrive having been told the opposite by a system the institution has no contract with and no visibility into. The published record was right. The representation circulating about it was not. And the potential consequences, which include a complaint stalling because it went to the wrong office, an escalation being deferred on wrong advice, or a policyholder misjudging their protection, together with a governance question about information customers may rely on, are the institution's to understand, evidence and manage, regardless of who typed the answer.

Conventional governance controls often focus on information the institution publishes or systems it operates directly, and here the published record is accurate. This risk sits outside the institution's controlled publishing perimeter, where consumers may consult public AI before approaching the institution or reading the official source.

The certified findings in this reading concern the steps a consumer takes when something has already gone wrong: when to escalate, where to complain, and what is protected. Those are the moments a misled consumer can least afford bad information. A longer escalation wait than the published process sets may defer redress. A complaint sent to the wrong office may not satisfy the prerequisite for referral. A protection benefit understated may leave a policyholder deciding on a worse picture than they actually have. These are the ordinary business of dealing with an insurer, which is why getting them wrong, invisibly, is worth a risk function's time.

How to govern AI answers about your institution

The answer is not to chase the AI, which cannot be corrected the way an institution's own website copy can, but to govern the surface with the same seriousness applied to any surface through which customers encounter claims about the institution. In practice that is a repeatable loop: watch what the major systems are telling consumers about how to complain, how long to wait, and what the schemes around you protect; check each answer against the official published fact, the regulator's complaint process, the scheme's own coverage page, your own complaints materials; separate the genuine misstatements from the false alarms before acting on either, and where a flag turns out to be correct, say so; work out why a real one drifted, since a borrowed waiting period, a wrong internal office and an understated benefit call for different fixes; rank what could actually mislead a consumer ahead of what is merely imprecise; and keep a dated record of what was said and when. Done steadily, an open-ended worry becomes a managed process. The institution finds out before the customer does, corrects the sources it controls, and can show its work to a board, or to a regulator, when asked what it is doing about AI.

Reading history

This page is refreshed in place each month at the same address. Earlier readings are demoted here rather than deleted, so the record stays visible. Each entry below is a standalone summary of that reading. We do not draw comparisons between readings.

First reading (June 2026). A scheme-only reading, checking public AI against PIDM's Takaful and Insurance Benefits Protection System (TIPS). Its certified findings concerned the scheme's limits and dates. Accurate handling was also present in the reviewed examples.

Second reading (July 2026). A reading covering insurer-level and scheme questions. Its certified findings concerned a total-permanent-disability benefit stated as ending at age 60 where the policy contract ran it to the insured's 64th birthday, a claim-acknowledgement standard against a published commitment of within 7 working days, and a complaint clock stated in calendar days where the insurer published working days. Accurate handling was also present in the reviewed examples.

Third reading (August 2026). This reading's certified findings are summarised above: complaint-escalation timing, which internal unit begins a complaint, and two points about what the public protection scheme covers.

We are not drawing a trend line across these readings. We do not yet have a stable, re-asked tracking panel with comparable provenance across readings, so each is a standalone single-month reading. We will consider trend language only when that panel exists.

Read on

This is an early read from an ongoing barometer Lawnise is building on how public AI answers questions about Malaysia's insurance sector, refreshed in place each month with more categories to follow over time. If you would like to see what public AI is currently saying about your own institution and the schemes around it, checked, the way these were, against the official published facts, we can scope a private AI answer baseline: sector-context, no obligation.