The Circular Dependency Problem
Tesla has deployed over 1,000 Optimus Gen 3 robots in production at Gigafactory Texas and Fremont. SoftBank acquired ABB Robotics for $5.375B, signaling mainstream commercialization of physical AI. Deloitte projects 80% of global companies will deploy physical AI by 2027.
Yet the insurance industry—the sector responsible for pricing and managing the liability of these robots—has only 7% enterprise-wide AI adoption success. This creates a structural paradox: the technology creating the risk and the institution managing that risk are on completely different adoption curves.
When an Optimus robot damages a product, injures a worker, or malfunctions in a factory, the traditional product liability framework faces novel questions: Is the manufacturer liable? The AI model provider? The factory operator who deployed the robot? And critically: the insurers writing these policies are using the same AI-powered claims adjudication tools (like Sedgwick's Sidekick) that raises these liability questions in the first place.
This is not a theoretical concern. It is an immediate governance gap with no clear resolution path.
The Insurer Capability Deficit
Insurance experts explicitly state they "do not believe insurance will reach fully autonomous AI decisions without human oversight—possibly ever." That statement reflects not a preference for human judgment but a recognition that AI-powered risk assessment systems themselves are too opaque to insure.
The SAS 2026 study quantifies the deficit:
- 50% of insurers cite data governance gaps—they cannot reliably collect, standardize, and analyze the data needed to assess physical AI liability.
- 44% report AI talent shortages—nobody at the insurer knows how to evaluate Optimus's failure modes or risk profile.
- Only 7% have scaled enterprise AI—meaning 93% of the insurance industry is pricing physical AI liability using pre-AI actuarial methods.
The skill gap is not modest. Pricing physical AI liability requires understanding failure modes of specific robotic systems, failure rates under different operating conditions, and the extent of human oversight in those conditions. That's actuarial work that requires either deep robotics expertise or access to manufacturers' failure data.
Manufacturers do not publish failure rates. Tesla does not disclose Optimus malfunction data. So insurers are left guessing at risk using historical product liability comparables—which do not apply to autonomous systems.
Tesla's Self-Insurance Strategy
Tesla's answer is implicit: self-insure. Deploy Optimus exclusively in owned factories where Tesla absorbs all the risk. This solves the immediate liability question for internal operations.
But Tesla's consumer-facing target is end-of-2027 deployment. Leasing Optimus robots to third-party factories requires external liability insurance. And that insurance market does not yet exist in a form that covers the risk competently.
The math: Tesla targets 10M Optimus units per year at dedicated Giga Texas factory, each with 10,000 components. If each robot causes injury or property damage once every 10,000 operating hours (a conservative assumption), the liability exposure is massive. Insurers cannot price exposure they do not understand.
The AI-Insurance Circular Dependency
Here is where the deeper problem emerges: insurers must use AI to assess AI-related liability, but they have not successfully deployed AI at scale themselves. This creates a credibility problem.
If an insurer uses a claims-processing agent (like Sedgwick's Sidekick) to decide whether a physical AI liability claim should be paid, what happens when the agent's decision is wrong? Who is liable? The insurer for deploying an unreliable agent? The model provider for the underlying AI? The policyholder for not contesting the decision?
This is not a hypothetical concern. If an insurer's AI agent denies a physical AI liability claim incorrectly, the policyholder sues the insurer. The insurer's defense is "our agent made the decision"—which is not a legal defense. The insurer is responsible for the agent's outputs, whether human or AI.
So insurers face a bind: they cannot assess physical AI risk without AI tools, but deploying AI tools creates new liability for the insurer itself. The 93% that have not scaled AI internally are choosing the safer path: avoid the complexity entirely and underwrite physical AI using conservative (and therefore inaccurate) pre-AI methods.
The Emergence of AI-Native Insurtech
A new category of AI-native physical AI insurers will emerge to fill this gap. These are companies that:
- Specialize exclusively in physical AI liability
- Deploy multi-model agentic AI to assess robotic failure modes
- Partner directly with robotics manufacturers for failure rate data
- Price policies at the level of specific robotic systems and operating conditions, not broad product liability categories
The addressable market is enormous. Goldman Sachs projects the humanoid robot market alone at $38B by 2035. Every deployed unit needs liability coverage. If traditional insurers are unprepared to write this coverage, startups filling the gap can capture significant market share.
The template exists: Sedgwick's Sidekick shows that agentic AI can handle complex claims workflows at scale. An AI-native insurer could build similar orchestration to assess physical AI risk, integrate manufacturer telemetry data, and price policies dynamically based on real-time failure rate updates.
Political Friction: The Robot Tax Complication
The UAW has proposed a "robot tax" on physical AI deployment. If enacted, this adds cost uncertainty to every physical AI deployment scenario and complicates actuarial calculations for insurers.
A robot tax changes the economics of humanoid robotics, which in turn changes the risk profile that insurers need to model. Robots become more expensive to deploy, so fewer get deployed, which lowers aggregate liability exposure. But the tax introduces a regulatory variable that traditional insurers are not equipped to model.
AI-native insurtech companies are better positioned to absorb this kind of regulatory complexity because they are not constrained by legacy actuarial frameworks. They can model scenarios (with/without robot tax) as part of policy pricing.
Timeline and Implications
The risk gap is immediate (Tesla's consumer Optimus launch is 2027). Traditional insurers will not solve the gap in time. The period from 2026-2028 will see third-party physical AI deployment operating with inadequate or no external liability insurance.
This creates two parallel markets:
- Internal deployment (Tesla factories): Self-insured. High risk. Acceptable for the manufacturer.
- Third-party deployment (other companies' factories): Either uninsured or underinsured via AI-native startups with limited track record.
By 2028-2030, market consolidation will occur. Either traditional insurers will acquire AI-native physical AI specialists, or the specialists will grow large enough to dominate. The companies that solve the circular dependency first (AI assessment of AI-related risk) win significant market share.
What This Means for Practitioners
If you are deploying physical AI in third-party settings: insurance availability and cost are not yet mature. Plan for self-insurance, partial coverage through AI-native startups, or deployment in owned facilities where you absorb the risk. Do not assume traditional insurance will adequately cover your exposure.
If you are building AI-native insurance products: physical AI liability is a greenfield market. Differentiate on AI assessment capability and manufacturer data partnerships, not on brand or scale.
If you are pricing physical AI liability in a traditional insurer: you are competing against uninsured deployers and AI-native specialists. Either accelerate your internal AI scaling, or outsource risk assessment to partners who have done so.