Who Pays When the AI Gets It Wrong? Inside the Race to Insure Artificial Intelligence

Artificial Intelligence Insurance: In 2018, Amazon discovered that an experimental recruiting tool had learned to penalise resumés containing the word “women’s”—including references like “Women’s Chess Club.” The company had no idea the algorithm had absorbed this bias from historical hiring data. It abandoned the tool. But the episode exposed a deeper problem: who bears the cost when AI systems make or influence decisions that create liability?

That question is now urgent. In Mobley v. Workday, the plaintiff alleged that Workday’s algorithmic hiring tools discriminated based on race, disability and age. The U.S. Equal Employment Opportunity Commission intervened, arguing that Workday could potentially face liability as an employment agency or agent. While the court later dismissed the claims based on the employment-agency theory, it allowed the plaintiff’s disparate-impact claims concerning algorithmic screening to proceed. Meta employees have alleged that AI-assisted systems played a role in discriminatory layoffs—though Meta disputes that AI made the termination decisions.

These are not edge cases. They reveal something more important than simple coverage gaps: a fundamental uncertainty about which insurance policies respond to AI-related losses, and how that coverage may change when new exclusions are introduced at renewal.

As enterprises across India’s financial services sector embed AI into underwriting, claims processing, hiring, and customer decisions, the insurance problem is not that AI losses are uninsured. It is that AI creates losses cutting across multiple insurance lines, while policy language, exclusions and accountability are evolving at different speeds.

Also Read: Tamil Nadu Health Insurance: Govt To Launch Plans For Elderly. All You Need To Know

Artificial Intelligence: The Real Problem: Uncertainty, Not an Uninsured Risk

The central challenge is not that AI is uninsured. Many traditional policies—employment practices liability, cyber, professional indemnity, D&O and general liability—may already respond to certain AI-related losses, depending on the underlying exposure and specific policy wording. The problem is uncertainty.

“You can delegate work to AI, but you can’t delegate accountability,” said Raj Goodman Anand, Founder, AI-First Mindset. “I describe AI as a very fast, very confident intern. It can do brilliant work, but somebody still needs to check it and own the final decision.”

The insurance question is sharper: which existing policy responds to the loss that occurred? And will that coverage survive renewal?

Artificial Intelligence: AI Exclusions Are Reshaping Policy Renewals

When insurance specialists began auditing renewal policies for clients, they found something alarming: insurers are now introducing AI-specific exclusions and tighter wording. “Businesses need to check,” Goodman advised. “Ask your broker, in writing, exactly where an AI-related loss would be covered. The risk is falling between your own insurance and the AI vendor’s contract.”

The concern is increasingly about what gets excluded at renewal, not whether something was covered initially. Jones Day’s 2026 analysis notes that many existing programmes may already respond to AI-related losses. But the absence of an AI exclusion does not guarantee coverage either. The question is whether the underlying policy was designed to respond to the particular loss, and whether any exclusion overrides that coverage.

Artificial Intelligence: Who Is Liable When the AI Stack Fails?

Insurers themselves are struggling with a fundamental problem: how do you price a risk when there is limited historical data, the technology evolves faster than actuarial models, and accountability lines remain legally murky?

“The honest answer is: whoever controlled the layer that failed,” said Nitish Gopalani, CEO, Fonada. “But most enterprises never have that conversation in writing before go-live. That’s where the real damage happens.”

Liability depends on which part of the AI stack failed. A useful analytical framework identifies several control points: the model and platform (technology provider), the data used to train and operate it (data owner), the configuration and deployment (enterprise), and human oversight (enterprise). A failure can originate at any point.

India’s RBI FREE-AI Committee report, published in August 2025, emphasises that regulated entities cannot outsource their responsibility for managing AI risks simply by outsourcing the technology itself. But the specific legal liability for a particular loss depends on contracts, causation, and what each party controlled.

Artificial Intelligence: Why the Absence of an Exclusion Doesn’t Mean Coverage

Yet many business leaders operate under a different assumption. “It’s very common to assume that because a policy does not have an explicit exclusion, it automatically absorbs algorithmic workflows,” said Atrey Bhardwaj, Chief Growth Officer of Probus. “That’s not how it works anymore.”

Sougata Basu, Founder & Chief Executive Officer, Cashrich, describes his operational philosophy bluntly: treat AI output like the work of a junior analyst—useful, fast, but never the final word. “Every fund that reaches a client must pass a suitability check, and the regulator will hold the company responsible for that check, not any AI model. So our rule is simple: AI can inform the work, but liability stays with a named person.” If you cannot name who approved and owns an outcome, Basu argues, you have a liability gap that no policy wording will fix.

Artificial Intelligence: Insurers Respond With New AI Exclusions and Endorsements

This principle—accountability follows control—is spreading through India’s BFSI sector. The insurance industry, however, is caught between adaptation and inertia. Traditional E&O policies were not necessarily drafted with autonomous or AI-assisted decision-making in mind, creating uncertainty around how existing definitions, exclusions and causation provisions apply. Some insurers are now introducing dedicated technology endorsements and AI-specific products. Others are introducing AI-specific exclusions. Verisk’s Insurance Services Office (ISO) has introduced optional generative-AI exclusion forms with 2026 edition dates, including CG 40 47, CG 40 48 and CG 35 08. Some insurers have begun drafting AI-specific carve-outs across liability lines.

“Insurance has always relied on looking backward—using decades of historical data to price future risk,” Bhardwaj noted. “With AI, that rearview mirror is mostly empty. The challenge for underwriters is shifting from being math historians to becoming forward-looking partners.”

Artificial Intelligence: Pricing AI Risk Without Historical Loss Data

The pricing challenge runs deeper. Concentration risk is particularly troubling: thousands of enterprises may rely on the same underlying AI systems, which means a single model failure can cascade into losses across an entire client base. With limited claims history and rapidly changing technology, insurers have less traditional loss data on which to rely. Underwriting is increasingly dependent on scenario analysis, evaluation of controls and governance, and expert judgment rather than historical loss patterns alone.

Alongside past claims experience, insurers are beginning to evaluate a company’s current data hygiene, testing protocols, governance structures, and human oversight mechanisms. “Risk assessment can no longer focus solely on technology performance. It must encompass how AI is governed, monitored, and integrated into business processes,” said Ritesh Varma, VP & Insurance Head, Newgen Software. “In many cases, the quality of oversight, transparency, and accountability may become as important as the technology itself.”

Also Read: Govt’s Bima Bharosa portal received over 800 insurance complaints every day in FY26

Artificial Intelligence: Five Risk Zones in Customer-Facing AI Decisions

The stakes are particularly high in applications involving customer-facing decisions. Advocate Ashwini Kumar, Founder of My Legal Expert (MLE), identifies five critical risk zones: inaccurate or biased outputs, inadequate human oversight, data privacy failures, lack of auditability, and the absence of clear contractual allocation of liability between businesses and their AI vendors.

“The ‘human-in-the-loop’ principle is crucial,” Kumar said. “Businesses need to establish who approved the deployment, what controls were in place, whether the system was tested, whether human review was available, and whether the decision can subsequently be explained and audited. If companies cannot answer these questions, no policy will protect them.”

Artificial Intelligence: Real-World Cases: Air Canada, Workday and Meta

Real disputes are already emerging. The Air Canada decision found the company responsible for negligent misrepresentation arising from incorrect information provided by its chatbot about bereavement fares. The Workday and Meta cases illustrate how questions around AI-assisted employment decisions are making their way through the courts. In Mobley v. Workday, the court allowed disparate-impact claims concerning algorithmic screening to proceed, while the EEOC had separately intervened in the case to argue that Workday could potentially face liability under an employment-agency theory. In the Meta case, employees have alleged AI tools played a role in layoff decisions—though Meta disputes that AI made the termination decisions.

These disputes point in the same direction: companies cannot assume that delegating a decision to an AI system automatically transfers legal responsibility to the technology.

The lesson is consistent across all these disputes: “The algorithm did it” does not make your legal responsibilities disappear. What these early cases reveal is a pattern of companies that rushed AI adoption without establishing clear lines of accountability. That is fundamentally a governance problem, not an insurance problem. Insurance can transfer some of the financial consequences, but it cannot substitute for governance.

Artificial Intelligence: What Businesses Should Do Before Deploying AI

For business leaders, the message is stark. Do not assume that the absence of an explicit exclusion means coverage exists. Do not rely on insurance to substitute for governance. And before deploying AI in any high-impact function, sit down with your insurance broker and legal team to map exactly which layer of control your business owns, which your vendor owns, and who pays if each layer fails.

There is one more trap to avoid: AI getting better does not mean it has become reliably right. In fact, the better it sounds, the easier it is for people to trust it too much. Speed and confidence are not the same as accountability. And having insurance is not the same as having that particular loss covered.

Efficiency and automation have their place. But as AI moves from being a productivity tool to a decision-making engine, accountability cannot be automated away. It can only be pushed around. The question is not whether someone pays when AI gets it wrong. The question is who—and whether they saw it coming.

Follow us on Twitter for latest updates