AI Errors Just Cost You $1M. Are You Covered?

"Do not make the mistake of assuming that your liability for AI errors is covered by your existing insurance. (McLane Middleton, March 2025)

‍

Summary:

• The focus on AI’s risks is shifting to performance failures
• Hallucinations, inaccuracies, bias and other unpredictable errors—are exposing gaps in traditional coverage. 
• Today’s policies are silent on AI risks and weren’t designed for probabilistic systems, giving rise to coverage risk. 
• A small claim might pass today—but will the same wording hold for a $100,000, $500,000 or $1M losses? 
• As incidents and lawsuits climb, more brokers and risk managers are choosing to get ahead of uncertainty, with new AI-specific coverage.

‍

Beyond Cyber Risks: Understanding AI Performance Failures

While AI-related cyber threats have dominated headlines and underwriting discussions, a parallel risk is quietly gaining momentum. AI performance failures—ranging from hallucinations to inaccurate outputs—are disrupting operations, exposing companies to liability, and revealing critical gaps in traditional insurance coverage.

One reason for these emerging risks is that AI model integrity and reliability are inherently difficult to assess. And unlike traditional software, AI models operate probabilistically—making predictions based on patterns in data–meaning their performance can vary significantly depending on inputs, subject matter, production conditions, and the rigor of developer training, testing, and validation. This means, an AI model may perform as intended and still experience performance failures, or have an inherent defect leading to losses that might not be covered under existing policies.

AI Performance Issues Set to Stress Test Traditional Coverage

Even the most advanced models fail at surprising rates, with OpenAI CEO Sam Altman recently acknowledging that the company’s newest model hallucinates more than one-third of the time (Futurism, March 2025). A Stanford paper pointed out last year that LLMs used by law firms were similarly unreliable. The challenge posed by AI model performance dovetail with broader trends around AI risk: 

• More than 60% of S&P 500 companies disclosed AI-related risks as material in their 10-K filings (WSJ, December 2024);
• AI incidents have surged 26-fold since 2012 (Stanford AI Index); and, 
• Over 140 AI-related lawsuits are currently pending in the U.S. AI Litigation Database).

Growing awareness of AI performance risks is exposing critical gaps in traditional coverage—particularly for companies increasingly reliant on AI technologies in core operations. As one broker put it to us: “Sure, a small claim might pass today—but will the same wording hold for a $100,000, $500,000 or $1M claim?“

‍

Some Concerns about Common Policies 

The probabilistic nature of AI performance often conflicts with traditional insurance structures designed for deterministic software. Here are the most common concerns we’ve heard from brokers working with risk managers to place coverage tailored to AI exposures.  

• Cyber:
  
  • Primarily covers external threats like cyberattacks or breaches, but not internal AI errors such as inaccuracies or hallucinations. 
  • Unclear whether coverage includes passive or non-breach leaks of sensitive or personally identifiable information (PII).  
• Technology Errors & Omissions (Tech E&O): 
  
  • Typically covers technology service failures or negligence explicitly as stipulated in contracts. 
  • Ambiguous coverage, as many vendors fail to define clear model performance benchmarks or KPIs for their AI solutions. 
  • AI underperformance might get classified as an inherent defect, creating coverage ambiguity.
  • Policies may not respond if an enterprise modifies or customizes the vendor’s AI model.
  • Vendor coverage limits may be insufficient in relation to potential exposure
  • Enterprises typically do not carry Tech E&O coverage.
• General Errors & Omissions (E&O): 
  
  • Designed for human errors and negligence, not probabilistic systems
  • Standard E&O is at odds with the known unpredictability of AI model performance, which is at odds with concepts of human error, negligence and traditional business tools.  
  • May not cover models developed internally

‍

Recent Observations About AI Performance Coverage Gaps

Industry experts are responding to growing concerns about traditional insurance policies falling short on AI performance risk. Here just a few of the most recent takes:

• How Air Canada's chatbot mishap uncovered a new exposure (Insurance Business America, January 2025)
• Does your professional liability insurance cover AI? Don’t be so sure. (ABA Journal February 2025)
• Businesses should consider affirmative AI policies specifically covering losses where AI technologies do not perform as intended. (Kennedys, February 2025)
• Do not make the mistake of assuming that your liability for AI errors is covered by your existing insurance. (McLane Middleton, March 2025)

‍

Action Steps for Brokers and Risk Managers

A growing number of brokers and risk managers are using specialized submission forms to review and assess AI-related exposures, with a focus on performance issues. A thorough review should include the following:

• Identify Priority AI Innovations: What AI solutions or tools have become critical to the organization over the past year?
• Understand the Technology Stack: What types of AI models—such as machine learning, large language models, or generative AI—are powering these innovations?
• Assess Evaluation and Testing Rigor: Have these models been tested for performance, robustness, and reliability under real-world conditions?
• Risk Quantification: What kinds of damages could result from AI failures, including errors, hallucinations, or misleading outputs?
• Evaluate Insurance Fit: Given these risks, how confident are we that traditional policies will respond – including for large claims?

‍

About us: At Armilla AI, we specialize in affirmative insurance solutions for AI performance and liability, backed by A-rated insurers like Chaucer, Axis Capital, Swiss Re, and Greenlight Re. Whether you’re evaluating AI risk, assessing coverage issues or looking for clear, AI-specific protection, we’re ready to support you in strengthening your approach to AI risk management and coverage. 

‍

Contact us to learn how Armilla AI helps enterprises assess, quantify, and insure AI risk in partnership with top brokerages.

Are you a broker or risk manager? Covering AI is Armilla's newsletter for insurance brokers and enterprise risk managers. It's free to subscribe.

‍