Robots can do the work. Can we insure it?

The gap between "can" and "will"

Anthropic's new report, What work can robots do?, puts two numbers side by side. Robots can already perform about 74% of physical job tasks in the US, roughly a third of all working hours. Yet they are cheaper than human labor for only 0.3% of tasks.

Most coverage has focused on that gap as a story about jobs and timelines. At Boop, we read it as a story about risk. Whether a robot gets deployed depends on what it costs to run, how much people trust it, and what regulators allow. Insurance touches all three.

Here are three findings we think matter most from an insurance perspective, and the questions they raise.

1. Insurance is already part of the automation math

When Anthropic estimates what it costs a robot to do a job, insurance is one of the line items. It sits alongside hardware, integration, maintenance, software, energy, and human oversight.

That matters because the margins are thin. For packers and packagers, the largest occupation where robots are already cost-competitive, the report estimates robots come out about $2,500 a year cheaper than a worker. Robotaxis are estimated at roughly $7,000 a year more than human drivers.

When the difference is that small, the price of risk can decide the outcome. If insurers can't price robot risk accurately, they price it defensively. Defensive premiums push robots back above cost parity. Better-priced risk does the opposite: it moves deployments that are almost viable into the viable column.

Insurance isn't just something that follows adoption. It is one of the levers that sets its pace.

2. Exposure tiers are also risk tiers

The report sorts tasks by the kind of environment a robot needs to do them:

Tier	Where the robot works	Example	Share of physical tasks
E1	Purpose-built spaces, like an assembly line	Packing meals on a conveyor	About 50%
E2	Structured human workplaces, like a warehouse or hospital	Delivering medications to nursing stations	About 22%
E3	Unstructured environments, like a city road	Driving passengers	About 2%

Anthropic built this scale to measure capability. Read from an insurer's seat, it also measures exposure to people. An E1 robot works behind fences, away from workers. An E3 robot shares the road with the public. Each step up means more people nearby, less predictable conditions, and more ways for an incident to happen.

We think E2 is where the most important risk questions sit right now. These are robots working alongside people in warehouses, hospitals, and stores. The category is large, it is growing, and the incident patterns there are still poorly understood. When a mobile robot and a worker share an aisle, who is liable for a collision: the operator, the integrator, or the manufacturer?

3. Trust is a barrier, and evidence is how it comes down

Cost isn't the only thing holding robots back. The report estimates that current regulations would block robots from about 14% of physical tasks, and that human preferences hold them back from about a quarter. People may not trust a robot with a task, or may simply prefer a person.

The authors also note that better robots could ease this resistance by addressing safety concerns. We agree, with one addition: being safer isn't enough. Safety has to be shown.

The report's own standard says a robot counts only if it matches humans on reliability, error rates, and speed. But for robots deployed in the real world, there is no shared way to record and compare how often things go wrong, how badly, and why. Each manufacturer and operator tracks incidents its own way, if at all.

That gap affects everyone. Regulators can't write evidence-based rules without comparable data. Customers can't judge whether a robot is safe enough to work next to their staff. Insurers can't price what they can't measure.

This is why Boop built the Robot Data Recorder. A common way of recording what happened, and under what conditions, is the foundation for fair pricing, sensible regulation, and public trust.

Questions this report raises

Anthropic's report is one of the most careful attempts yet to map where robots can work today. It also leaves open questions that the robotics and insurance industries need to answer together:

• What should a robot incident record contain? What do operators, manufacturers, regulators, and insurers each need to know after something goes wrong?
• Where does liability land in shared spaces? As robots move into warehouses and hospitals, how is responsibility split between the people who build, integrate, and operate them?
• Who carries the risk when work moves from people to machines? The most robot-exposed jobs involve heavy lifting, extreme heat, and hazardous materials. As robots take on that work, some injury risk shifts from workers to machines. How should coverage follow it?

Each of these questions comes back to the same need: reliable, comparable incident data. That's the problem our Robot Data Recorder was designed to solve. If you build, deploy, or insure robots, we'd be glad to show you how it works. Reach the Boop team at info@trustboop.com, or read the RDR white paper.

Sources

• Legate-Yang and Massenkoff, What work can robots do?, Anthropic, September 30, 2026