🚀AIB #06 - AI Agents Governance risks, Improved Hurrican Forecasts, ChatGPT Health, UK Insurers Top Priorities

Hey it’s Fabio here,

Quick behind-the-scenes: we’ve spent the last few weekends redesigning the site and building a set of AI agents that now run most of the engine for AI Insurer Brief.
If you reply this email with YES, I will be happy to break down exactly what the agents do, how they’re set up, etc...

Now to this week’s signals.

This week, Experian’s breach forecast is blunt: AI agents are becoming the next attack surface - poor agent governance could trigger the next wave of incidents.
Brokers are getting serious too: BIBA and Markel are building an AI Academy to move adoption from curiosity to execution. Gallagher Re called 2025 the start of a “new era” in hurricane forecasting as agencies use AI to spot storms days earlier.

Outside insurance, “personal intelligence” is being embedded into everyday tools, and new models are getting better at predicting human behaviour

Here’s what caught my eye this week (in less than 4 min) 👇

Gallagher Re says the 2025 hurricane season signals a “new era” in meteorology: official agencies are now using AI models alongside traditional forecasting to improve how storms are detected and tracked. The practical benefit for insurance is clearer near-term visibility - earlier signals on storm formation and likely path, which helps operators plan response and manage exposure with more confidence.

Gallagher Re notes these data-driven AI models have increased confidence in longer-range Atlantic hurricane track forecasts by multiple days. However, the report is clear on limitations: AI has been less reliable at predicting other loss-driving factors such as intensity and precipitation. So the message is “better directional guidance,” not perfect loss forecasting.

Google DeepMind’s tropical cyclone model, launched in June 2025 via Google Weather Lab, is the most visible example. It produces ensemble forecasts up to 15 days out, offering earlier probabilistic views of potential scenarios.

Why it matters

Dun & Bradstreet’s latest research suggests UK insurers are heading into 2026 with two priorities running in parallel: scale AI internally while tightening operational risk management, especially around third parties, fraud, and external exposures.

The message is simple: both goals rise or fall on data quality and governance.

The survey highlights that 37% of UK insurers rank internal AI adoption as a top 2026 priority. The same 37% put third-party risk reduction at the top.

Dun & Bradstreet reads this as a deliberate balancing act - insurers want efficiency and growth from AI, while recognising that more outsourcing, ecosystems, and partners increase exposure to operational failures, cyber incidents, and conduct risk.

Other priorities appear less dominant: 31% focus on market expansion, and 26% cite embedded finance initiatives and improved decision-making.

The biggest constraint is data. Dun & Bradstreet reports that more than half of insurers have seen AI initiatives fail due to poor data quality, and only 17% currently run unified data platforms. Fragmented systems and legacy infrastructure also amplify operational risk, from fraud to cyber to supply chain disruption.

Unsurprisingly, 76% of respondents say operational risk is a top concern.

Why it matters: AI ROI will come from operators who fix data foundations and third-party controls first - otherwise AI spend turns into repeat pilot failures and higher risk.

BIBA and Markel are partnering to launch an AI Academy for BIBA members, building on their well-received 2025 Guide to AI. The goal is to move brokers from basic awareness to practical, confident adoption—covering both opportunities (efficiency, better service) and risks (safe and appropriate use).

The Academy will be designed to deepen brokers’ understanding of where AI can genuinely help their businesses and how to deploy it responsibly. BIBA’s CEO, Graeme Trudgill, frames AI as a fast-moving capability that most firms will find useful in some form—primarily to drive efficiencies and free time for higher-value customer service. Markel’s Simon Johnston reinforces that AI is already an active topic among brokers, and that the missing piece is often shared understanding: what the technology is, where it fits, and how to use it safely in day-to-day broking work.

The AI Academy is currently in development and is expected to open to broker “students” early in 2026.

Why it matters: Brokers are the front door of the UK insurance market. If brokers become more AI-capable, the entire value chain benefits - faster submissions, cleaner data capture, quicker servicing, and better customer responsiveness. This also sets a baseline for “safe AI” adoption across the market.

Experian warns: AI agents are becoming the new attack surface

Experian’s 2026 data breach forecast puts AI at the centre of how cyber risk is evolving- specifically the scope, frequency, and cost of incidents.
The key message is that AI isn’t just a defensive tool anymore. It’s becoming an accelerant for attackers, changing both the methods and the scale of cybercrime.

Experian highlights a new threat vector: agentic AI systems.
The concern is that sophisticated attackers could inject their own agents to disrupt or hijack the orchestration and governance of a victim’s agents.
In practical terms, that could mean operational disruption, theft of money or data, or ransomware-style outcomes - except executed through compromised automation rather than a traditional manual attack chain.
Experian goes further, suggesting AI agents may overtake human error as a leading cause of breaches, because the automation layer itself becomes a primary target.

The forecast also warns that AI can enable faster data extraction and help criminals “stitch together” richer identity profiles from stolen fragments across sources - driving a potential spike in identity theft.
On the malware side, Experian expects AI to support more advanced mutating malicious code (polymorphic/metamorphic malware), making detection and containment harder.

Finally, Experian connects this to quantum computing: it argues quantum is already driving increased attacks and could eventually break current encryption. While AI security tools help today, Experian expects breach-prevention firms to accelerate “quantum-ready” capabilities to stay ahead.

Why it matters: Insurers and brokers are becoming more exposed as they adopt AI agents for ops and customer service. This increases the attack surface and raises expectations for controls: strong agent governance, access management, logging, third-party oversight, and incident response readiness—before automation is scaled.

See you next Wednesday! 👊

Share the AI Insurer Brief. Stack your perks.

How to unlock it: