AI Underwriting Gets Real in Cyber and Terrorism

A senior war and terrorism underwriter hire in London and a cyber underwriter appointment tied to a new product launch don’t look like “AI news” at first glance. But they are.

These People Moves—Markel International appointing James Howell as senior war and terrorism underwriter, and Tokio Marine Kiln (TMK) appointing Olivia Jackson as a cyber underwriter—signal where underwriting is heading in 2026: humans who can translate fast-changing risk into decisions, backed by AI models that can keep up with the data.

I’m opinionated on this: most insurance teams don’t have an “AI problem.” They have an execution problem. The winners aren’t the ones with the flashiest model. They’re the ones pairing the right underwriting talent with the right AI workflow—especially in complex lines like terrorism/political violence and cyber.

Why these hires matter for AI in insurance

These appointments matter because war/terrorism and cyber are two of the clearest examples of underwriting where traditional “look-back” methods break down.

• In war and terrorism, exposure can change in hours. Aggregations, exclusions, and contract certainty become the product.
• In cyber, the insured’s risk posture can change every time they add a vendor, migrate cloud workloads, or patch late.

So why hire senior specialists now? Because AI-driven underwriting is raising the bar. Underwriters are expected to:

1. Make decisions faster without lowering quality
2. Explain pricing and terms more clearly to brokers and insureds
3. Handle more granular segmentation (not broad “industry buckets”)
4. Keep portfolios within appetite despite volatility

AI helps, but only if underwriting leadership can operationalize it. That’s what these roles are really about.

War and terrorism underwriting: where AI adds value (and where it can’t)

AI is useful in war and terrorism underwriting when it improves exposure awareness, aggregation control, and scenario-based decisioning—not when it pretends to “predict the next event.”

Markel’s hire of Howell—nearly a decade in terrorism and political violence underwriting plus early exposure analysis work—highlights a practical truth: the best AI use cases here start with exposure discipline.

The core AI use cases in terrorism and political violence

Here’s what strong AI underwriting support looks like in this line:

• Entity resolution and accumulation mapping: Matching insured names, locations, and subsidiaries across messy broker submissions to understand true concentration.
• Geospatial enrichment: Normalizing addresses and layering in hazard proximity, urban density, critical infrastructure adjacency, and event-history context.
• Scenario modeling at portfolio scale: Running “what if” losses (by trigger, location, and insured class) to keep accumulations inside tolerance.
• Broker submission triage: Using natural language processing to route submissions to the right underwriter, flag missing details, and standardize intake.

A good one-liner for your team: AI doesn’t replace judgement in terrorism underwriting; it replaces blind spots.

What AI shouldn’t be doing in war/terrorism

Teams get burned when they try to over-automate judgement calls that must remain human-led. A few red flags:

• Treating model outputs like truth rather than inputs
• Ignoring contract wording and relying on “risk scores” alone
• Using black-box models without portfolio and claims feedback loops

In practice, war and terrorism underwriting is as much about terms, triggers, and aggregation control as it is about probability. Senior underwriting leadership (like Howell’s role is designed for) is what keeps AI outputs grounded in reality.

A practical workflow that works

If you’re building an AI-assisted workflow for war/terrorism, aim for this order:

1. Submission intake standardization (structured data + document extraction)
2. Exposure mapping (locations, insured entities, limits, attachment points)
3. Accumulation checks (portfolio + syndicate/capital view)
4. Scenario tests (define triggers; run stress cases)
5. Decision + documentation (underwriter narrative + model support)

That last step is underrated. Regulators, brokers, and internal audit don’t want a number—they want a story.

Cyber underwriting: AI is now part of the product

Cyber is different. AI isn’t only an underwriting tool—it’s also embedded in the insured’s risk and in the insurer’s operations.

TMK’s appointment of Jackson comes alongside the rollout of flagship offerings (TMK Cyber Ctrl and Enterprise Ctrl) and additional capability hires. That combination points to a market reality: cyber underwriting is moving toward continuous risk assessment, not a once-a-year renewal conversation.

The most effective AI applications in cyber underwriting

Cyber carriers and MGAs are leaning into a handful of proven AI patterns:

• Control-based scoring: Turning security signals (patch cadence, MFA coverage, endpoint controls, backups) into underwriting factors.
• External attack surface monitoring: Tracking exposed services, leaked credentials, and misconfigurations to identify drift over time.
• Claims routing and triage: Classifying incidents (ransomware vs. BEC vs. vendor outage) and escalating based on severity patterns.
• Fraud and anomaly detection: Flagging suspicious claim timing, repeated indicators across insureds, or inconsistent incident narratives.

A stance I’ll defend: cyber is becoming “underwrite-the-controls,” and AI is the only scalable way to do that consistently.

What changes for brokers and insureds in 2026

If you’re a broker or risk manager, expect these shifts to become more common:

• More structured underwriting questionnaires (less narrative-only)
• More evidence requests (logs, control attestations, vendor lists)
• More mid-term engagement (not just renewal season)
• More differentiated pricing between “good controls” and “paper controls”

That’s not carriers being difficult. It’s the market responding to a harsh truth: cyber loss drivers are operational, and operational drift is constant.

The hidden signal in “People Moves”: underwriting leadership is the AI adoption engine

A lot of AI in insurance fails because it’s treated as an IT initiative. Underwriting leaders then get a new dashboard and carry on as usual.

These hires suggest the opposite approach: put experienced underwriters in the center, then modernize the workflows around them.

What senior underwriters do that models can’t

In both war/terrorism and cyber, experienced underwriters excel at:

• Knowing which data is “nice to have” vs. deal-critical
• Negotiating terms that reduce ambiguity (and claims disputes)
• Building broker relationships that improve submission quality
• Teaching teams consistent decision frameworks

That’s why Markel’s description emphasizes Howell’s role in training and maintaining underwriting standards. AI outputs are only as good as the underwriting discipline around them.

The operating model that scales AI underwriting

If your team is trying to scale AI-driven underwriting, copy this structure:

• Underwriting owns the decisioning logic (appetite, segmentation, referrals)
• Data/AI owns the measurement system (model monitoring, drift, feedback)
• Claims owns the reality check (loss drivers, wording gaps, cycle time)
• Actuarial owns portfolio economics (rate adequacy, tail risk, capital)

When those four groups share the same playbook, AI becomes practical—not performative.

What to do next: a 30-day plan to make AI underwriting real

If you’re responsible for underwriting operations, product, or innovation, here’s a tight plan that generates measurable progress without boiling the ocean.

Week 1: Pick one workflow, not “AI”

Choose a single underwriting pain point, such as:

• cyber submission triage and missing-info follow-up
• war/terrorism accumulation identification across insured entities
• referral rules automation for specific attachment points

Write down the current baseline: cycle time, touchpoints, and error rates.

Week 2: Define “good data” and “good decisions”

For your selected workflow, define:

• Minimum data required to quote
• The top 10 reasons for decline or referral
• The 5 fields that are most often wrong or missing

This becomes your model and process specification. Without it, you’ll automate chaos.

Week 3: Build a human-in-the-loop pilot

Run a pilot where AI supports a decision, but doesn’t make it. Examples:

• auto-summarize submissions into a standardized underwriting brief
• suggest appetite fit and referral rationale
• flag accumulation conflicts before quoting

Track two metrics only: cycle time reduction and underwriter override rate (and why they overrode).

Week 4: Close the loop with claims and brokers

Meet with claims to validate that your underwriting signals match real loss drivers. Then meet with 3–5 top brokers and show them what “good submissions” look like.

One of the fastest wins I’ve seen: give brokers a one-page “submission quality scorecard.” It improves data quality, reduces rework, and makes AI outputs more reliable.

Where this goes next for AI in insurance

AI in insurance is shifting from experimentation to accountability. In 2026, carriers will be judged less by whether they “use AI” and more by whether they can prove three things:

• Their underwriting decisions are consistent
• Their models are monitored and explainable
• Their portfolios stay inside risk appetite as conditions change

People Moves like these are a signal that the market understands it. Complex risks require specialized underwriting leadership—and that leadership increasingly depends on AI-driven underwriting tools to keep pace.

If you’re building capabilities in cyber insurance underwriting or war and terrorism risk modeling, the question to ask internally isn’t “Do we have an AI strategy?” It’s “Which underwriting decisions will we make better next quarter—and how will we measure it?”