Industry news roundup

CBUAE publishes AI governance guidance for all UAE financial institutions. The Central Bank issued its Guidance Note on Consumer Protection and Responsible Adoption of AI and Machine Learning on 23 February 2026, establishing governance expectations for all licensed financial institutions, including every insurer operating in the UAE. The guidance is principles-based rather than prescriptive, but its inclusion in the CBUAE Rulebook signals clear regulatory expectations. It covers board-level accountability for AI outcomes, fairness and non-discrimination requirements, transparency and explainability standards, and effective human oversight across the AI lifecycle. Institutions that ignore it will find themselves explaining that choice to the regulator. (CBUAE Rulebook, February 2026)

Emiratization penalties hit AED 108,000 per missing employee. Since January 2026, UAE private sector companies with more than 50 employees face a penalty of AED 108,000 for each Emirati employee below target. The insurance sector's current Emiratization rate sits at approximately 22%, against a target of 30% by end of 2026. Companies that fail to pay face work permit freezes. For insurers already managing compressed margins, this creates a structural cost pressure that makes the economics of outsourcing non-core functions harder to ignore. (MOHRE / Gulf News, January 2026)

MIC Global launches GCC expansion from Qatar via QIC partnership. US-based insurtech MIC Global secured Qatar regulatory approval and deepened its strategic partnership with Qatar Insurance Company to launch MiIncome, an embedded income-protection product distributed through banks, remittance platforms, and telecom providers. Qatar serves as the regional hub, with sequential expansion planned for Kuwait, Oman, and Saudi Arabia. The model validates embedded distribution as a viable entry strategy for GCC insurance markets, and signals that large regional carriers like QIC are actively partnering with insurtechs rather than building in-house. (PR Newswire, 20 February 2026)

Point of view: The AI governance circular is an opportunity, not a checkbox

The CBUAE's AI guidance note changes the governance conversation for every insurer in the UAE. The guidance is principles-based, not prescriptive. But it sits in the CBUAE Rulebook, covers board-level accountability for AI outcomes, and sets expectations on fairness testing, transparency, and human oversight. No insurer that ignores it will have an easy conversation with the regulator.

Many UAE insurers have invested in AI tools over the past three years. Fraud detection models, claims triage automation, pricing algorithms, chatbots. The investment is typically driven by efficiency targets and competitive pressure, which make sense. What is missing in nearly every case is a governance framework around those tools. The AI was deployed; the oversight was deferred.

That deferral is now a governance gap the regulator has explicitly flagged. The CBUAE's guidance expects documented governance proportionate to institutional size, which means every insurer now needs to answer three questions it probably has not asked: 1) What AI models are running in production? 2) Who is accountable for their outputs? And 3) when was the last time anyone tested whether those outputs were fair?

The instinct will be to treat this as a compliance exercise: assign it to risk, produce a governance policy, file it alongside the AML framework. That approach satisfies the letter of the regulation and misses the point.

The insurers that treat this circular as an opportunity will build genuine AI competence. A proper AI inventory forces an honest assessment of what's actually working and what's just running. Annual bias testing produces data that improves model performance, not just data that satisfies an auditor. Board-level accountability creates a feedback loop between the people making strategic AI investment decisions and the people managing operational AI risk. None of that happens if governance is treated as paperwork.

Most GCC insurers do not have in-house AI governance expertise. The guidance is principles-based rather than prescriptive, which gives flexibility but also creates ambiguity. Insurers that have been running AI models without formal oversight cannot build a governance framework overnight, and consultants who have never operated inside a claims workflow will produce governance documents that look rigorous and miss the operational reality.

The gap is between the regulation and the capability to comply with it. Insurers that close that gap fastest will not just avoid enforcement risk. They will build the institutional muscle to deploy AI responsibly at scale, which is a competitive advantage that compounds over time.

Operational efficiency spotlight: Building resilience through cash flow control and operational flexibility

In uncertain times, cash is king. In uncertain times, operational flexibility is king. Both statements are true, and for GCC motor insurers navigating volatile conditions, they point to the same imperative: reduce the operational cost base without reducing capability.

The two largest controllable cost lines in any motor insurance operation are claims cost and personnel spend. Both can be reduced materially without degrading service, but only if the insurer rethinks how work gets done rather than simply cutting headcount or squeezing repair bills.

Rethinking the operating model

Claims cost containment starts with process discipline, not negotiation. An insurer that benchmarks every repair job against thousands of comparable claims, routes work through governed workshop networks, and applies standardized quality gates catches cost leakage structurally. That containment holds under pressure because it is embedded in how claims are processed, not layered on as a quarterly initiative that erodes when attention shifts.

Personnel spend responds to the same structural logic. A claims operation built on fixed headcount carries idle capacity when volumes drop and scrambles when volumes rise. Restructuring the execution layer through specialist operating partners converts that fixed cost into a variable one. The insurer pays for throughput, not seats. When volume rises, capacity rises with it. When it contracts, cost contracts proportionally.

Where AI fits: support, not replacement

Artificial intelligence accelerates both levers, but it does not replace the need for sound processes and capable people. Document parsing reduces manual data entry. Anomaly detection flags cost patterns a human reviewer would miss. Predictive routing matches claims to the right handler faster. These are real efficiency gains, but they require well-designed processes to operate within and experienced professionals to oversee.

The market is full of vendors positioning AI as the solution to everything. That framing misses the point. Axxion's view is that processes and people come first, and AI serves as a strong support, but never as a lead. An AI tool deployed into a broken process automates the breakage. The same tool deployed into a well-structured operation compounds every efficiency gain the process already delivers.

The insurers building real resilience are not chasing the latest AI announcement. They are tightening operating models, converting fixed costs to variable costs, embedding containment controls into the claims chain, and using AI to amplify what already works. That combination of cash flow discipline and operational flexibility keeps an insurer performing when market conditions refuse to cooperate.

AI in insurance: Why deterministic guardrails matter more than model selection

The CBUAE's new AI governance guidance expects "effective human oversight" across the AI lifecycle. In practice, that means insurers need to demonstrate that their AI systems cannot produce outputs that violate business rules, regulatory requirements, or policy terms. For claims operations, this is not a theoretical concern. A language model that hallucinates a policy coverage determination or misclassifies a liability assessment creates real financial and regulatory exposure.

The solution is not to avoid large language models. It is to wrap them in deterministic programming that constrains what they can and cannot do.

How this works in practice: FNOL as example

Consider a first notification of loss intake workflow. A policyholder reports an accident through a digital channel. The submission includes free-text descriptions, photographs, and uploaded documents. An LLM is useful here: it can extract structured information from unstructured text, classify damage types from images, and pre-populate claim fields.

Without guardrails, the LLM might populate a coverage field based on its training data rather than the actual policy terms. It might assign a liability determination based on the narrative description without cross-referencing the police report. It might classify a total loss based on visual damage assessment without checking the vehicle's insured value against repair cost thresholds.

Deterministic guardrails prevent all of these failures. The architecture works in layers. The LLM handles extraction and classification, tasks it performs well. A rules engine then validates every populated field against the policy database, regulatory requirements, and business rules. Coverage fields are checked against the actual policy schedule, not inferred. Liability determinations require corroborating evidence from at least two sources. Total loss thresholds are calculated from the insured value and repair estimate, not from visual assessment alone.

The LLM never makes a decision. It proposes; the deterministic layer validates. When validation fails, the claim is routed to a human reviewer with the specific field flagged and the reason for the flag documented. This creates an audit trail that satisfies the CBUAE's governance requirements while preserving the efficiency benefits of AI-assisted processing.

The insurance industry's skepticism toward AI is not irrational. Models hallucinate. Training data contains biases. Outputs are probabilistic, not deterministic. The answer is not to reject AI or to deploy it without constraints. It is to build systems where AI handles what it is good at and deterministic rules handle what must be right. The CBUAE's governance guidance makes this architectural choice the expected standard. Smart operators were already building this way.

Market data and benchmarks

Introducing the Axxion Claims Data Index

<aside>

1.3 million vehicle-years of earned exposure. Over 315,000 individual claims. Data from UAE-licensed insurers spanning every major vehicle segment, repair channel, and manufacturer origin.

</aside>

Starting with this edition, The Claim Curve will feature a regular data section drawing on Axxion's proprietary datasets. This edition's benchmark: what EVs actually cost to insure compared to equivalent ICE vehicles.

The EV repair cost gap: Tesla vs. German premium ICE

A Tesla costs 2.2 times as much to insure on a pure-risk basis as a comparably priced BMW, Mercedes, or Audi. That is not an estimate. It is what 6,361 vehicle-years of Tesla exposure and 2,163 claims show when compared to 22,206 vehicle-years of German premium ICE (BMW 3 Series/X3, Mercedes C-Class/GLC, Audi A4/A5/Q5). Both sides carry full actuarial credibility.

The severity premium (+38%) compounds with a frequency premium (+62%) to produce a burning cost gap of 123%. An insurer with 600 Tesla policies priced at the German premium ICE benchmark is accumulating roughly AED 1.5 million per year in pure-risk undercharge. That gap does not surface in a quarterly loss ratio review until it has already compounded across several reporting periods, because the Tesla book is small relative to the total portfolio.

The pattern holds beyond Tesla. Across the full EV segment (excluding Lexus hybrids used as taxi/ride-hail vehicles, which distort the data), EVs carry a burning cost of AED 3,798 against AED 1,666 for the ICE fleet. The severity premium is 111%. The frequency premium is just 8%. This is a repair cost problem, not a driving behavior problem.

Source: Axxion burning cost dataset, 1,257,717 vehicle-years of earned exposure, 315,712 claims, UAE-licensed motor insurers. Credibility assessed using the limited fluctuation method at 95% confidence.

From the desk of Axxion

Axxion's MD has been writing about operational resilience during periods of regional instability on LinkedIn, covering how insurance operations can maintain service continuity when external conditions are unpredictable. The series is available on Frederik Bisbjerg's LinkedIn profile.

This is the first edition of The Claim Curve, and it will only be as useful as the topics it covers. What questions should the next edition answer? What data would be most valuable? Reply to this email or connect on LinkedIn. The editorial direction belongs to the readers as much as the team.

Thank you for reading so far - until next time, Frederik

‍