AI Underwriting Talent Shifts in Cyber and Terrorism

Specialty insurance is where underwriting skill gets tested hardest—and where AI in insurance underwriting is starting to matter most. When a carrier hires a senior underwriter for war and terrorism, or adds a cyber underwriter right as a new cyber product line ships, they’re not “just filling a seat.” They’re buying better judgment, sharper broker relationships, and a faster feedback loop between risk signals and pricing.

That’s why two London market moves from this week are worth paying attention to: Markel International appointing James Howell as Senior War and Terrorism Underwriter, and Tokio Marine Kiln (TMK) bringing in Olivia Jackson as a Cyber Underwriter. On the surface, this is a classic people-moves story. Underneath, it’s a signal of where the market is investing: complex risks that demand data-driven underwriting, stronger modeling, and tighter controls.

Here’s the stance I’ll take: AI won’t replace specialty underwriters—but it will replace underwriters and teams who don’t know how to work with AI. These appointments make sense because cyber and political violence are two lines where the volume, volatility, and ambiguity are too high for manual processes alone.

Why these hires matter for AI-driven underwriting

These appointments matter because they align people, product strategy, and analytics at the exact moment specialty lines are becoming harder to underwrite with “traditional” playbooks.

War/terrorism and cyber share three traits that make AI-driven underwriting unusually valuable:

• High signal-to-noise: open-source intelligence, threat chatter, vulnerability data, sanctions changes, geopolitics, vendor outages, and more.
• Fast-changing exposure: a risk can materially shift inside a policy term.
• Non-linear loss: when things go wrong, they can go very wrong—quickly.

An experienced underwriter brings pattern recognition and market instincts. AI brings scale: it can continuously scan, triage, and summarize risk signals. The winning operating model is a partnership:

Underwriters set the questions; AI speeds up the answers; governance keeps it honest.

These people moves also highlight a practical truth carriers sometimes avoid saying out loud: AI projects fail when they’re “tech-led” instead of “underwriting-led.” Appointing leaders who can mentor teams, shape underwriting guidelines, and collaborate across claims/actuarial/product is exactly how you keep AI grounded in commercial reality.

Markel’s war and terrorism appointment: where AI actually helps

Markel’s appointment of James Howell as Senior War and Terrorism Underwriter (based in London, reporting to Chloe Gordge) lands in a market described as “pivotal,” with rising geopolitical tensions driving more complex threats and higher client demand.

The war/terrorism underwriting problem isn’t data scarcity—it’s relevance

War and terrorism underwriting has plenty of data. The real issue is filtering it into what matters for a specific insured, location, supply chain, and coverage form.

AI contributes most in three places:

1. Exposure mapping and aggregation

   • Using entity resolution to link insured names, subsidiaries, addresses, and asset locations.
   • Monitoring accumulation (for example, multiple insureds in the same district or near critical infrastructure).

2. Threat signal triage

   • Summarizing open-source reports and structured feeds into underwriter-ready briefings.
   • Flagging changes that should trigger re-assessment (e.g., sanctions changes, border escalations, increased attack frequency).

3. Policy wording consistency and claims defensibility

   • Comparing endorsements and exclusions against internal guidelines.
   • Highlighting wording drift across brokers/regions that can create silent exposure.

If you’re building an “AI stack” for political violence, don’t start with a flashy model. Start with underwriting workflow.

A practical workflow I’ve seen work: AI as the first-pass analyst

A realistic operating model for war/terrorism underwriting looks like this:

• AI pre-briefs each submission with a 1-page risk synopsis (locations, proximate hotspots, relevant events in the last 90 days, key exclusions/endorsements)
• Underwriter reviews and adjusts what matters (risk appetite, required controls, pricing posture)
• Second-line governance (actuarial and claims) validates that assumptions and wordings match portfolio strategy

That last step matters because political violence losses often turn into wording disputes. If your AI tooling improves speed but reduces consistency, you haven’t improved underwriting—you’ve increased future friction.

TMK’s cyber underwriting hire: the “controls era” is here

TMK’s appointment of Olivia Jackson as Cyber Underwriter follows the launch of TMK Cyber Ctrl and Enterprise Ctrl, and comes alongside other capability-building hires (including cyber risk analysis and reinsurance solutions).

This is the part of cyber underwriting I feel strongly about: cyber is moving from “questionnaire underwriting” to “evidence underwriting.” And that shift is inseparable from AI.

Why cyber underwriting is becoming AI-assisted by default

Cyber submissions are notoriously inconsistent. Controls are described in different terms across insureds, regions, and broker decks. Security posture changes mid-term. And incident frequency isn’t evenly distributed.

AI helps because it can convert messy inputs into structured, comparable fields—without forcing every broker and insured into the same template.

Examples of what AI-enabled cyber underwriting can do well:

• Normalize controls language into a consistent taxonomy (MFA, EDR, backups, patch cadence, privileged access)
• Detect missing “must-have” controls for certain classes
• Identify concentration risk by cloud provider, MSP, or key vendor
• Summarize external signals (attack surface, leaked credentials indicators, domain misconfigurations) into a risk score the underwriter can challenge

The underwriting posture shift: price less on “industry,” more on “control proof”

Cyber products branded around “Ctrl” are essentially making a promise: controls matter, and they’ll be evaluated.

That raises a question many carriers are now facing: How do we verify controls without turning underwriting into a six-week audit?

The strongest approach is tiered verification:

1. Self-attestation for smaller limits or low-complexity risks
2. Evidence checks (screenshots, configuration exports, limited scans) for mid-market
3. Continuous monitoring and claims-ready documentation for large limits and complex enterprises

AI supports all three tiers by automating document extraction, summarization, and consistency checks.

What these moves reveal about the London market heading into 2026

Late December is when many teams are closing renewals, cleaning up portfolios, and setting next-year priorities. Hiring in war/terrorism and cyber right now is a clear message: specialty risk is where carriers expect both growth and volatility.

Here are three trends these appointments reinforce.

1) Underwriting leadership is becoming product-and-data leadership

The Markel role explicitly includes underwriting policy development, cross-functional collaboration, and training/mentoring. That’s not accidental.

When AI enters underwriting, you need leaders who can:

• Translate appetite into rules and exceptions
• Decide what AI outputs are actionable vs. noise
• Create repeatable training so judgment scales beyond one person

If your “AI underwriting initiative” doesn’t include underwriter enablement and training, it’s a pilot that never leaves the lab.

2) The broker relationship is evolving into a data relationship

Both roles emphasize broker partnerships. That’s where AI adoption gets real.

As carriers push for evidence-based underwriting (especially in cyber), brokers who can deliver structured data and credible control proof will win faster turnaround and better outcomes for clients. Carriers that can ingest broker data cleanly—via AI extraction and workflow automation—will quote faster and more consistently.

3) Model risk management is becoming a frontline requirement

AI models can drift. Data feeds can be wrong. Summaries can miss nuance. Specialty lines are not forgiving when errors stack.

If you’re deploying AI for underwriting in cyber or political violence, you need explicit guardrails:

• Documented model purpose and limitations
• Human-in-the-loop decisions for material changes (limits, exclusions, declinations)
• Audit trails for why key decisions were made
• Regular back-testing against loss experience and near-miss events

This is where underwriters like Howell and Jackson become critical—because the best governance is practical. It fits the pace of the desk.

If you’re building AI in underwriting, copy these playbooks

You don’t need to be a global specialty carrier to apply what’s happening here. If you’re an insurer, MGA, or broker building AI in insurance capabilities, these are the steps that actually produce results.

Build around decisions, not dashboards

Start by mapping the 5–10 underwriting decisions that drive loss ratio and growth.

For war/terrorism, that might be:

• Accept/decline thresholds by geography
• Sub-limits, deductibles, and key exclusions
• Accumulation controls

For cyber, that might be:

• Minimum controls for eligibility
• Required endorsements for certain classes
• Limit and retention recommendations by control maturity

Then build AI assistance that directly supports those decisions.

Use AI to reduce cycle time where it matters

Quoting speed isn’t about being “fast” everywhere. It’s about being fast on the boring parts so experts can focus on the hard parts.

High-value automation targets:

• Submission intake and triage
• Document extraction and control mapping
• Wording comparison against guidelines
• Aggregation reporting

Make underwriters co-owners of the AI system

If underwriters only see AI outputs at the end, adoption will be shallow and skepticism will be permanent.

A better approach:

• Have underwriters label what “good” looks like (examples of strong controls, unacceptable exposures, common broker phrasing)
• Create feedback buttons inside workflow (“useful / not useful” with a reason)
• Review false positives weekly, not quarterly

This is how AI becomes a tool the desk trusts.

The lead-generation reality: buyers want confidence, not buzzwords

If you sell into carriers, MGAs, or brokers, these two hires highlight the buying mindset you’ll encounter in 2026: they’re paying for confidence under uncertainty. That means vendors and internal teams need to show measurable outcomes.

The outcomes that get budget approval in specialty underwriting tend to be:

• Shorter quote turnaround time (measured in hours/days)
• Improved consistency of underwriting decisions across teams
• Better visibility into accumulation and aggregation
• Clearer claims defensibility through documented rationale and wording controls

If your AI solution can’t tie to at least two of those, it’s not “strategic.” It’s optional.

As this AI in Insurance series keeps showing, the most durable AI wins come from pairing technology with domain leaders who can operationalize it. Markel and TMK are signaling exactly that.

If you’re planning your 2026 underwriting roadmap, here’s a useful gut-check: Do your underwriters spend more time thinking about risk, or moving data around to describe risk? Your answer tells you where AI belongs next.