AI Underwriting Talent Shifts in Cyber & War Risk

Late December is when many insurers quietly set their 2026 chessboard—budget lines get finalized, teams get reshuffled, and specialty portfolios get sharper. Two London market hires reported this week look like routine “people moves” on the surface: Markel International appointing James Howell as Senior War and Terrorism Underwriter, and Tokio Marine Kiln appointing Olivia Jackson as a Cyber Underwriter.

Most companies get this wrong: they treat these announcements as HR news. They’re not. In specialty lines, senior underwriting hires are strategy signals. They tell you where capacity is headed, which risks are getting more complex, and—crucially for this AI in Insurance series—where carriers think data-driven underwriting will separate winners from everyone else.

War/terrorism and cyber aren’t just “hot” classes. They’re the two areas where underwriting judgment gets stress-tested by fast-changing threat patterns, thin historical loss data, and heavy reliance on external intelligence. That’s exactly the environment where AI underwriting tools can help, and exactly why these roles increasingly demand underwriters who are as comfortable with analytics as they are with broker relationships.

People moves are strategy moves (especially in specialty)

The simplest way to read these appointments: specialty underwriting is becoming more technical, more data-dependent, and harder to scale without AI. When an insurer adds senior talent in war/terrorism or cyber, it’s often because the portfolio needs one (or more) of the following:

• Sharper risk selection as claims severity rises
• Faster quoting and renewal triage across crowded submission pipelines
• Better portfolio steering (aggregation, clash, accumulation)
• More credible broker conversations grounded in evidence, not vibes

This matters because specialty portfolios are where small pricing errors become outsized P&L swings. If you misread aggregation in political violence, you can stack correlated exposures without realizing it. If you underprice a cyber layer because you trusted a checklist, you’ll learn the hard way when ransomware turns into business interruption, regulatory response, and class action defense.

AI doesn’t replace the underwriter in these lines. It raises the floor: it standardizes what “good” looks like, flags what humans miss, and helps teams act consistently under time pressure.

War and terrorism underwriting: why AI matters when history doesn’t repeat

Markel International’s appointment of James Howell (joining from AXIS, with prior exposure analysis experience) lands at a moment when war and terrorism risks are being reshaped by geopolitical volatility and more ambiguous threat actors. Even when you don’t name specific events, you can feel the underwriting reality: uncertainty is up, and brokers want answers faster.

The underwriting problem: fast-changing threats + aggregation risk

War/terrorism underwriting has three chronic constraints:

1. Sparse, noisy historical data
2. High correlation (many insureds impacted by the same trigger)
3. Policy wording sensitivity (what exactly is excluded/covered?)

Underwriters often rely on scenario thinking: What if X happens in region Y? That’s rational—but manual scenario work can’t keep pace with modern exposure complexity.

Where AI underwriting tools actually help

A practical AI approach in war/terrorism isn’t about predicting the next attack. It’s about improving underwriting execution:

• Exposure enrichment: match insured locations to geospatial risk layers, critical infrastructure proximity, and event footprints.
• Portfolio accumulation monitoring: detect concentration by region/industry/supply chain node and show “clash” in near real time.
• Broker submission triage: classify submissions, extract key fields from PDFs, and route edge cases to senior review.
• Wording and endorsement analytics: use NLP to compare clauses, highlight differences, and reduce silent coverage creep at renewal.

One snippet-worthy truth: In specialty lines, AI’s biggest ROI is reducing unforced errors.

If Howell’s remit includes underwriting policy development, training, and broker relationships, that’s a strong hint Markel wants repeatable, teachable underwriting decisions—the kind you can encode into guidelines, decision support, and portfolio guardrails.

A simple “AI + human” workflow for political violence

Here’s what I’ve found works when teams want AI support without turning underwriting into a black box:

1. AI extracts and normalizes exposure data (locations, values, operations)
2. Models generate scenario loss ranges and accumulation alerts
3. Underwriter decides: accept/decline, price, apply sublimits, refine wording
4. Post-bind monitoring: watch accumulation drift and update risk appetite

The underwriter stays accountable. AI handles the repetitive and the combinatorial.

Cyber underwriting: why every carrier is chasing better signal

Tokio Marine Kiln’s hire of Olivia Jackson into its Cyber & Enterprise Risk team follows the launch of its cyber offerings (including TMK’s “Ctrl” products). That sequence matters. Product launches create a predictable operational squeeze: submissions rise, brokers test appetite boundaries, and renewal discipline gets harder.

Cyber is also where AI is already changing daily work. Not theoretically—operationally.

The underwriting problem: cyber risk changes faster than policy cycles

Cyber underwriting is plagued by a gap between how fast threats evolve and how slowly most insurance processes move.

A policy renews annually. A threat actor changes tactics weekly.

So the underwriting job becomes a constant fight for fresh, decision-grade data:

• External attack surface and exposed services
• MFA and privileged access controls
• Backup posture and recovery testing
• Third-party/vendor dependency
• Prior incidents and near misses

Where AI improves cyber risk selection (and where it doesn’t)

AI can absolutely improve underwriting speed and consistency in cyber—but only if it’s aimed at the right targets.

High-value applications:

• Submission intelligence: extract controls, revenue, industry, claims history; detect contradictions.
• Control efficacy scoring: combine questionnaire answers with external telemetry to reduce misrepresentation risk.
• Loss drivers and segmentation: cluster insureds by control patterns and predict frequency/severity bands.
• Early warning for portfolio shifts: detect deterioration (e.g., more exposed RDP-like services) and adjust appetite.

Low-value (often risky) applications:

• Fully automated “yes/no” decisions without underwriter review
• Opaque scoring that can’t be explained to brokers or regulators
• Models trained on biased or outdated incident data

A line you can reuse internally: If you can’t explain the model’s decision in plain English, you can’t trust it with capacity.

Why this hire signals “portfolio building,” not just replacement

Jackson’s background includes managing renewals, pursuing new business across territories, and cultivating broker partnerships. That profile is useful when a carrier wants to:

• Expand distribution without losing underwriting discipline
• Build a cyber book with consistent pricing logic across regions
• Align underwriting with product positioning (primary vs excess, industry focus, attachment points)

AI supports this by making underwriting standards scalable. But scaling cyber without strong underwriting leadership is how carriers end up with “growth” that’s really just accumulation of hidden downside.

The new underwriting profile: broker-first and data-literate

Both hires underline a broader market reality: modern specialty underwriting is a hybrid job.

The best underwriters still do the human parts extremely well:

• They earn broker trust.
• They negotiate clean terms.
• They spot when a risk story doesn’t add up.

But the job now also includes technical expectations that used to live elsewhere:

• Comfort with analytics dashboards and portfolio metrics
• Ability to pressure-test model outputs
• Understanding of data quality (missing fields, inconsistent exposure units)
• Familiarity with automation (triage, prefill, doc ingestion)

Here’s the stance I’ll take: If your underwriting leaders can’t work with data, your AI underwriting program will stall. Not because AI fails, but because the organization can’t translate model output into underwriting action.

A practical AI underwriting roadmap for specialty teams (90 days)

If you’re an insurer, MGA, or specialty team leader trying to turn “AI in underwriting” into measurable outcomes, this is a realistic 90-day plan that doesn’t require a full platform rebuild.

1) Pick one workflow where speed and accuracy both matter

Good starting points:

• Cyber renewals triage (what needs senior review?)
• Political violence accumulation monitoring
• Submission intake and data extraction

Bad starting points:

• “Automate all underwriting decisions”

2) Build a minimum viable data layer

You need:

• A consistent submission schema (even if incomplete)
• A way to track quote-to-bind and reasons for decline
• A simple feedback loop: underwriter overrides and why

3) Define success metrics that underwriting actually respects

Examples:

• Reduce time-to-quote by 30% for low-complexity submissions
• Increase “data-complete” submissions from 40% to 70%
• Cut referral volume by 20% without increasing loss ratio volatility

4) Put governance where it belongs: in underwriting leadership

AI programs fail when they’re owned entirely by IT or “innovation.” Specialty teams need:

• Named underwriting owners for each model/workflow
• Documented guidelines for when AI can auto-route vs must refer
• Audit trails for decisions (especially in cyber)

Common questions teams ask about AI underwriting in cyber and terrorism

Can AI price war/terrorism risk if events are unpredictable?

AI can’t predict specific events reliably, but it can systematize scenario analysis, improve exposure quality, and monitor accumulation—three things that directly improve pricing discipline.

Does AI reduce the need for senior underwriters?

No. It reduces avoidable manual work and improves consistency. Senior underwriters become more valuable because they spend more time on edge cases, wording, and portfolio steering.

What’s the fastest win in cyber underwriting?

Automating submission ingestion and triage (including control extraction and referral rules) is usually the quickest operational ROI—because it removes bottlenecks without changing risk appetite overnight.

What these hires suggest for 2026: specialty underwriting gets more automated—and more human

The Markel and Tokio Marine Kiln appointments point in the same direction: specialty underwriting is becoming a discipline where human judgment is still the differentiator, but AI-enabled process is the enabler.

If you’re building or buying AI underwriting capabilities, don’t start by asking which model is “best.” Start by asking which decisions you want to make faster, which errors you want to stop repeating, and which parts of underwriting you want your top talent to spend their time on.

For the rest of this AI in Insurance series, I’ll keep coming back to this: AI works in insurance when it respects the underwriting chain of responsibility. The organizations that get that right will scale specialty portfolios with fewer surprises.

If you’re planning your 2026 underwriting operating model, where would AI remove the most friction in your team—cyber submissions, accumulation control, wording review, or renewal triage?