Risks associated with AI currently face a coverage gap, Armilla AI CEO Karthik Ramakrishnan told The Insurer TV on the sidelines of this year's FERMA Forum.
Ramakrishnan noted that AI-related risks do not generally fall under traditional cyber policies, while it remains ambiguous as to whether Errors & Omissions (E&O) policies would cover AI risks.
“Cyber insurance relates to system breaches, business disruption and leaks of sensitive information. That doesn’t necessarily cover issues with AI, because there's no breach event that we’re looking at,” he said.
“E&O could cover AI. The problem is it's silent on those covers. And what we have found with the prospects that we speak to, is that they have a hard time getting an affirmative answer on whether AI is included or not. It's definitely not excluded, but it’s not clear if it’s affirmatively included either,” said Ramakrishnan.
Ramakrishnan’s comments echoed those of Philippe Cotelle, head of insurance risk management at Airbus Defence & Space and chair of FERMA’s digital committee, who told The Insurer TV: “We urge risk managers to review their contracts to see if potential damages related to AI are covered. Few insurance contracts account for AI risks.”
With McKinsey estimating that around 70 percent of companies will use AI by 2030, the lack of AI coverage will become a pressing issue. Ramakrishnan noted that the absence of clear AI coverage leaves companies exposed to potential liabilities.
“There are about 140 to 150 active lawsuits in the US related to AI, involving copyright, bias, discrimination and misuse,” he said. Ramakrishnan also cited a case in which Air Canada’s chatbot provided bad information, leading to a lawsuit that was settled in the consumer’s favour.
Armilla AI
This lack of coverage clarity and growing liability issues around AI led to a strategic pivot for Armilla AI. Originally a risk quantification company for AI systems, the firm is now an MGA focusing on AI insurance, with capacity from Greenlight Re, Chaucer and Swiss Re. The primary product that Armilla AI offers is a product guarantee for AI.
“So while you could be covered [for AI], companies don’t want to operate in the grey zone. And would like a black and white answer to that, which is what we're trying to answer,” explained Ramakrishnan.
“We have a current product in market, which is a product guarantee, where based on our assessment, we can guarantee the performance of an AI model that it will continue to perform, and if it doesn’t, then we have coverages that we can provide around cost recovery or damages that ensue,” added Ramakrishnan.
Armilla AI can offer AI insurance as it stress tests AI models by inputting unusual data.
“The most important part to test is actually the robustness of a model,” he said.
“What I mean by robustness is we actually run tests where we’ll stress test the model on data that it's not been trained on, and see how the model behaves in this out of distribution data. And if it handles those situations more elegantly, then it’s a less risky model, versus a model that loses its mind when it sees data that it’s not seen before,” said Ramakrishnan.
He noted that established insurers have not adopted similar AI risk modelling practices and rely more on a rule-of-thumb approach.
“The insurance industry’s AI knowledge base is still developing. That’s the gap we fill,” Ramakrishnan concluded.
Watch the full video interview to learn more about where coverage for AI risks is falling short and the innovation being done to find and provide workable solutions.
Additionally, we were pleased to be part of the consultative process that led to this highly informative policy note on the EU AI Act in partnership with FERMA.
