The rapid advance of so-called “physical artificial intelligence” - spanning autonomous vehicles and humanoid robotics - is poised to be transformational for the insurance sector. For one industry observer, this transformation could eclipse the impact of generative or agentic AI.
According to the International Data Corporation, worldwide spending on artificial intelligence is projected to surpass $500 billion by 2027, with a growing share directed towards physical applications such as robotics and autonomous mobility.
Meanwhile, Goldman Sachs estimates that the humanoid robotics market alone could reach $150 billion by 2035, driven by labour shortages and advances in machine learning.
Chris Raimondo (pictured), insurance consulting leader for the Americas at EY, believes physical AI will trigger a shift in liability models, underwriting practices and product design. But he said that insurers are only beginning to grapple with the implications of machines acting in the physical world.
“Physical AI could potentially be more disruptive to the insurance industry as a whole than digital AI, which is where most of the activity is taking place right now as insurers implement different AI capabilities and tools in their operations,” said Raimondo in an interview with Insurance Business.“We are starting to see insurers consider the implications that physical AI may have on their entire business models.”
At the heart of the disruption, he said, is a fundamental reallocation of liability. Traditional insurance models, particularly in motor and workers’ compensation lines, have long been anchored to human liability. Autonomous systems, however, distribute risk across a more complex value chain.
“The major trend will be a shift from pricing, evaluating, and managing risk around human operators to system-centric operators or system-centric liabilities,” Raimondo said. “Taking autonomous vehicles as the simplest example, personal and commercial auto insurance have traditionally been priced based on the driver. In the future, liability could look materially different.
“Liability may extend across multiple parties, including the hardware vendor (the vehicle manufacturer), the AI software platform powering the vehicle, and the company or individual that owns the vehicle… whether personal or part of a commercial fleet.”
This shift is already emerging as self-driving technologies scale. Data from the US National Highway Traffic Safety Administration shows advanced driver-assistance and autonomous systems are being deployed across millions of vehicles, with several manufacturers reporting billions of autonomous miles logged in testing and limited commercial operations.
As adoption grows, insurers face the prospect of underwriting risks tied not only to drivers, but also to “software versioning, sensor capabilities, and data generated by the vehicle,” said Raimondo.
“More broadly, products are becoming more dynamic, with pricing and underwriting continuously adjusted based on real-time data from AI-enabled systems,” he added.
Carriers are beginning to experiment with new models to address this emerging risk. For example, insurtech Lemonade this year launched a usage-based policy which adjusts pricing depending on whether a vehicle is in autonomous or human-driven mode.
Such products signal a broader shift towards dynamic underwriting, in which premiums are continuously recalibrated based on real-time data streams from vehicles and devices.
Beyond insuring the robots or vehicles themselves, physical AI also creates opportunities to insure the infrastructure that supports it. Raimondo cited growing interest in covering both digital and physical infrastructure (i.e. charging networks, data centers, smart roads and edge computing facilities) required for these technologies.
The transformation also extends beyond underwriting into claims processing. Today’s claims models rely heavily on human investigation and adjudication. Physical AI, by contrast, will generate vast quantities of machine data.
“Today, claims are investigated through established methods, but in the future, they will rely more heavily on sensor data, OEM logs, and software diagnostics. Evaluating software faults will become a core component of claims adjudication,” Raimondo noted.
While generative and agentic AI have dominated recent headlines, their impact on the insurance industry has largely been internal, focused on efficiency gains in areas such as underwriting, customer service and fraud detection. Physical AI, by contrast, strikes at the core of what insurers cover.
“It’s not just products and underwriting that will change,” Raimondo said, “the entire insurance value chain will need to adapt to support these new risks.
“We are still in the early stages. Adoption and investment are increasing, and insurers are beginning to recognize that their business models, not just their products, may need to evolve significantly.”
The pace of change will depend on adoption curves for autonomous vehicles and robotics, which remain uneven across regions. However, as investment accelerates and regulatory frameworks gradually evolve, insurers, brokers and other stakeholders face mounting pressure to adapt.
