As artificial intelligence agents take on more business tasks with less human oversight, insurers are starting to respond in very different ways. Some are creating policies to cover losses linked to AI errors, while others are adding clauses that block such claims altogether. The shift reflects growing concern over how companies are using agentic AI to handle decisions, workflows and client services on their own.
According to a report by Agence France-Presse, the insurance industry is now being pushed to rethink how traditional liability coverage applies to AI-driven mistakes.
Why Are Insurers Changing Their Approach To AI Risks?
Phil Dawson of specialist insurer Armilla said the purpose of advanced AI is often to replace a significant part of human support and supervision in decision-making. He said this creates pressure on the logic behind existing insurance products.
Businesses are adopting AI agents to perform computer-based tasks independently, even as the technology still produces errors such as hallucinations, where false information is presented confidently.
A research paper by analyst Sonal Madhok and law professor Anat Lior, published by brokerage firm Willis Towers late last year, said AI-related liability has so far often been absorbed under existing policies through what is known as silent coverage.
They argued that this may not continue for long and that insurers are likely to start addressing AI directly in policy language.
What Types Of AI Coverage Are Being Offered Now?
Jonathan Mitchell of Founder Shield said some insurers have moved past a wait-and-see approach. He noted that certain standard policies now carry absolute AI exclusion clauses, while others are being adjusted to include AI malfunction and hallucination risks. Founder Shield also offers extensions for real-world damage, such as AI over-ordering inventory.
Armilla tests AI models before offering coverage and may reject areas such as medical diagnostics or mental health applications.
Munich Re also provides AI-related coverage and says errors cannot be fully eliminated because AI models carry statistical uncertainty. As Agence France-Presse noted, Deloitte estimates the global AI insurance premium market could reach $4.8 billion by 2032.
