Week 21 · 18–24 May 2026

Three angles this week

3 angles · 45 items reviewed · generated Sun 17 May

P&C carriers are pricing cyber insurance on a model that no…

Observation

Anthropic's Mythos model autonomously found thousands of zero-day vulnerabilities across major operating systems, prompting the Treasury Secretary to convene bank CEOs in an emergency session.

Angle

P&C carriers are pricing cyber insurance on a model that no longer exists. The assumption that offensive capability requires scarce human expertise is gone. Every actuarial table built before Mythos is now a liability, not an asset.

Implication for P&C carriers

P&C carriers writing cyber lines need to reprice immediately. The frequency and severity assumptions underpinning current policy structures were built for a pre-Mythos threat landscape. This is not an incremental adjustment.

1 source · Exponential View

P&C carriers are about to gain access to enterprise-grade…

Observation

Chinese AI labs, constrained by chip export controls, are extracting 4–7x more intelligence per unit of compute than naive scaling would predict, while pricing equivalent-capability models at 10–28x less than US counterparts.

Angle

P&C carriers are about to gain access to enterprise-grade AI at a fraction of current costs. The efficiency race China was forced into is now a pricing weapon that will structurally compress what carriers pay for AI inference — and what they can charge for AI-enabled services.

Implication for P&C carriers

Carrier technology budgets built around current AI inference costs will look overstated within 18 months. The more urgent question is whether procurement and vendor governance frameworks can handle models from non-US jurisdictions given regulatory and data residency constraints.

3 sources · Exponential View +2 more

The 'deploy AI and see what happens' era is over.

Observation

OpenAI, Anthropic, and Google are each deploying armies of forward-deployed engineers into enterprises to physically rewire business processes around AI, with OpenAI committing $4 billion to its Deployment Company and Google hiring hundreds of new field engineers.

Angle

The 'deploy AI and see what happens' era is over. The model labs have concluded that AI value capture requires embedding people inside customers to redesign workflows top-down — exactly the pattern of mainframe-era computing. P&C carriers waiting for AI to self-deploy are misreading the moment.

Implication for P&C carriers

For P&C carriers, this means AI transformation is not a technology project. It is an organisational redesign effort that requires executive mandate, data readiness work, and sustained human-led implementation — more like a Guidewire delivery than a SaaS rollout.

2 sources · Stratechery +1 more