Mosaic has partnered with Munich Re to launch insurance cover for artificial intelligence models that make mistakes or underperform.

The Bermudian-based firm said it has partnered with Munich Re’s aiSure to offer up to €15 million (about $16 million) in coverage for financial losses arising from defined AI performance failures. The product uses parametric-style automatic triggers, rather than when there is a cyberattack or system outage.

“If AI tools don’t deliver on what they promised, someone absorbs that loss,” said Dennis Bertram, head of AI underwriting at Mosaic. “The commercial question is: what happens when the technology doesn’t deliver?”

AI systems are increasingly used to detect fraud, assess creditworthiness, screen job applicants and generate written content, according to the company. Even well-built models can produce incorrect outputs, sometimes referred to as “hallucinations”, or fail to meet accuracy targets. Mosaic said that in some cases, losses like undetected fraud or legal liabilities can be more than the cost of the software itself.

Munich Re has previously warned that AI-generated content can expose companies to intellectual property lawsuits, with US statutory damages reaching up to $150,000 per wilful infringement, according to a company whitepaper.

The launch comes as the BMA consults on the responsible use of artificial intelligence in Bermuda’s financial services sector. In a recent discussion paper, the regulator warned that poorly designed AI models could amplify discrimination, create cybersecurity vulnerabilities or lead to “automation bias”, where firms overly rely on machine outputs without enough oversight. The BMA has called for stronger governance, transparency and board-level understanding of AI risks.

Michael von Gablenz, head of Insure AI at Munich Re, said: “Our collaborative approach will enable us to support even more companies on their journey towards AI implementation.”

A PwC study cited by Munich Re show 98 per cent of chief executives see immediate benefits from AI, but 65 per cent report low or moderate trust in deploying it.