Insurers and reinsurers are starting to back a new class of cover aimed at AI models themselves, opening a niche that could help organizations manage the risks of relying on automated decision‑making, according to Paulo Salomao (pictured), national lead of strategy and consulting at Accenture Canada.
“There are some companies that are starting to emerge now that are looking to insure AI models,” Salomao said in an interview with Insurance Business. “I think that’s a very exciting place to be.”
He said those providers are appearing in Canada as well as in the US.
“I’m seeing them in Canada,” he said. “I was talking to one of them last week.” The idea is they are looking to insure the model itself and to ensure that the model is behaving per the specifications that it was designed to behave, he explained.
Salomao described the basic problem as one of performance risk. Organizations are beginning to embed AI models in critical processes, from customer service to underwriting, but cannot always be sure the models will behave as intended once exposed to real‑world data and use cases. That creates both technical and business exposure if the system underperforms or fails in a way that harms customers or operations.
He said the products now emerging tend to have two layers: one focused on fixing the model and another on compensating for the damage.
Salomao said a typical structure starts with cover for the model’s performance itself. If an AI system fails to meet the effectiveness level it was designed for, the policy would provide funds to rebuild or retrain it until it reaches the agreed standard, he noted.
The more novel element, he said, is coupling that with what he called business‑impact insurance: “So if the model doesn’t behave, I will help you fix it, but I will also indemnify you for the damage that it’s done to your organization.”
That damage could take a variety of forms depending on how the model is used, from operational disruption to customer remediation or regulatory issues if outputs are found to be biased or misleading. By tying the cover both to remediation costs and downstream business consequences, he said, the policies aim to create a more comprehensive response to model failure than traditional tech E&O or cyber insurance.
Salomao said reinsurers are showing interest in backing these products.
“There’s real appetite for reinsurers to do that,” he said. “It’s just a new source of risk diversification.”
He noted that the capital ultimately comes from reinsurance markets.
For insurers, he said, the concept is directly relevant as they increase their own use of AI in underwriting, claims and customer service. If core decisions are being shaped by models, then failures in those models can generate both financial and reputational loss – the kind of exposure that can, in principle, be laid off to a third party.
“If you think about an insurance company making underwriting decisions using an AI model, if the insurance companies can have reinsurance for that model and to ensure that model is operating the way it should work, that creates an interesting way to one, diversify risk exposure, and two, really accelerate the journey towards more autonomous operations,” he said.
Salomao said the availability of model‑level protection could make boards and executives more comfortable with deploying agentic AI deeper into their organizations. If part of the risk that a model misbehaves can be transferred, he said, that may lower the internal hurdle rates for using AI to automate or augment more tasks.
He framed the development of model insurance as part of a broader shift in how organizations think about AI risk management. Instead of only investing in preventive controls, monitoring and internal governance, they can now also consider risk transfer mechanisms that resemble traditional insurance and reinsurance structures, but applied to algorithms rather than physical assets.
The field remains new, with only “some companies” active and product designs still evolving, he said. But he argued that insuring AI models is likely to grow alongside increased reliance on those systems in financial services and other data‑intensive sectors.
