Insurers and reinsurers are starting to offer coverage specifically for mistakes made by artificial intelligence (AI) screening tools used in the lending industry. This new type of insurance focuses on protecting lenders from losses when AI models wrongly predict borrower risk, leading to higher-than-expected defaults.
The coverage targets what are called “excess errors” — cases where borrowers default more often than the AI model had forecasted due to mistakes in the model itself, rather than economic factors like interest rates or market changes. This approach helps separate the risks linked to AI errors from the usual underwriting and market uncertainties.
One company leading the way is MKIII, a start-up with just 11 employees that provides AI screening services to credit unions and community banks. MKIII has introduced insurance against model mistakes as part of its package. This insurance has helped lenders lower the capital they must hold against loans, which can free up funds to issue more loans. Bryan Adler, co-founder of MKIII, explained that while most decisions are automated, a staff member still reviews borderline cases for creditworthiness to ensure accuracy.
Major players in the reinsurance market are also getting involved. Munich Re has agreed to cover risks if their AI models fail. Greenlight Re, an alternative reinsurer, is backing these efforts as well. Another start-up, Armilla, that specializes in AI insurance, has assessed MKIII’s software and worked to secure coverage from reinsurers including Greenlight.
The insurance pay-outs kick in if borrower defaults surpass what the AI predicts solely due to errors in the model. Karthik Ramakrishnan, CEO of Armilla, noted that the coverage is tied directly to measurable mistakes in AI performance.
Pricing these policies requires understanding that all AI models carry a chance of error. Michael von Gablenz, an AI expert at Munich Re, pointed out that because AI is inherently probabilistic, mistakes cannot be fully avoided. Premiums vary depending on the expected error rate, covering a wide range from low to high.
MKIII said that lenders using its platform have collectively paid millions in premiums to secure tens of millions in coverage. This safety net enables them to write hundreds of millions of dollars worth of home loans. The insurance is purposely narrow in scope to just cover AI errors, leaving out broader economic or regulatory risks.
Still, some insurers remain cautious about offering AI-related coverage. Certain companies have even asked regulators in the US to exclude AI-related losses from existing policies. They worry that such claims could be large, interconnected, and hard to predict with traditional insurance terms.
For both insurers and brokers, this new wave of AI-focused insurance points to a shift toward more precise products that pay out based on specific model performance, rather than the usual indemnity-based claims. It also raises bigger questions about managing risk and regulatory oversight as AI tools become more common in financial services and beyond.
Overall, these developments highlight the insurance industry’s effort to adapt to the unique challenges posed by AI, aiming to support lenders while managing the uncertainties of machine-driven decision-making.