Insurance follows the machine — and right now it can't keep up

When people list what unlocks autonomous machines, insurance rarely appears — and that is the honest read. The major adoption frameworks put public trust, technical safety, viable business models, regulation, and cross-industry collaboration at the centre. The World Economic Forum and BCG's 2025 work on autonomous vehicles is built around those fronts; insurance is not one of them. Insurance follows the technology. It does not lead it.

But there is a real bottleneck inside insurance, and it is specific. The one task these frameworks hand to insurers is to build risk-assessment models — and the stated precondition for doing that well is large-scale data for robust models, which insurers today do not have. The constraint is not appetite, and it is not capital. It is evidence.

The deeper problem is that traditional pricing lags real safety badly. The Casualty Actuarial Society has shown that a vehicle which halves its losses earns only about an 8% discount after four years, and even a near-crashless vehicle earns roughly 15%. Annual, demographic-based pricing was never built to track a risk profile that can change with a single software update. When risk moves faster than the rails that price it, the price stops carrying information.

Meanwhile the safety evidence has arrived. Swiss Re's peer-reviewed analysis of 25.3 million autonomous Waymo miles found around 90% fewer injury claims and 88% fewer property-damage claims than human benchmarks over comparable routes. The result is geofenced and the injury sample is small, but the direction is clear: the risk on these machines is genuinely different, and it is still being priced on instruments designed for human drivers.

Underneath all of this, the premium pool is being re-pooled rather than shrunk. Autonomy moves loss off the human driver and onto the machine, the software, and the operator — out of personal and commercial motor lines and into product, professional, and cyber liability. Deloitte estimates that even a 20% premium shift would move about US$7 billion a year out of US commercial auto alone; Goldman Sachs projects robotaxi-related insurance above US$400 billion by 2035; KPMG expects product liability to account for roughly 57% of auto losses by 2050. These are US-centric projections, but the shape is consistent: the base does not disappear, it reforms around new lines and new data.

So insurance's role in adoption is narrower and sturdier than the usual slogan. It is not the thing that enables autonomy; it is a follower whose ability to keep up is gated by data. In practice, the machines that scale first will be the ones whose risk can be read continuously and explained to the capital standing behind the policy — not the ones with the best demo.

That is the layer the market is missing: a live, legible risk signal, per asset and per trip, that licensed insurers and reinsurers can actually underwrite from. The need is most acute in commercial fleets, robotaxis, and delivery — and in APAC markets the incumbent models barely touch — where cover is mandatory from day one and risk concentrates on a solvent operator that has to insure. YAS reads machine telemetry and produces that signal; the licensed insurer prices and carries the risk. The honest claim is not that insurance unlocks autonomy. It is that autonomy is reshaping one of the largest premium bases on earth, the old rails cannot see it clearly, and the constraint is data — which is exactly the gap worth closing.

Sources: World Economic Forum and BCG, autonomous-vehicle adoption framework (2025); Casualty Actuarial Society, Automated Vehicles Task Force (2018); Swiss Re and Waymo, peer-reviewed analysis in Traffic Injury Prevention (2024); Deloitte; Goldman Sachs (2025); KPMG.