AIG, WR Berkley, and Great American have filed with U.S. regulators seeking permission to offer policies that exclude claims arising from businesses using AI tools, including chatbots and autonomous agents. The filings, reviewed by the Financial Times, indicate growing caution among insurers as companies rush to adopt AI, despite high-profile mistakes and costly errors caused by “hallucinations” or flawed model outputs. WR Berkley proposed exclusions barring claims tied to “any actual or alleged use” of AI, encompassing products or services incorporating the technology. AIG told the Financial Times that while it had filed generative AI exclusions, it currently has no plans to implement them, leaving the option open for the future.
Industry experts cited by the Financial Times warn that AI outputs are opaque and unpredictable, making liability difficult to determine. Even insurers that offer coverage for AI-enhanced software, such as Mosaic at Lloyd’s of London, have avoided underwriting risks associated with large language models like ChatGPT. Legal and regulatory uncertainty adds to the challenge, as multiple parties—from developers to end users—could be held responsible if AI systems fail. High-profile cases illustrate the stakes: Wolf River Electric sued Google for at least $110 million over defamatory AI-generated statements, Air Canada was ordered to honor false discounts from a chatbot, and the UK engineering group Arup lost $25 million to fraudsters using a digitally cloned executive.
Some insurers have introduced targeted endorsements to clarify coverage for AI risks. QBE, for example, extended coverage for fines under the EU AI Act but capped payouts at 2.5% of the total policy limit, a move other companies have since mirrored. Chubb has agreed to cover limited AI risks while excluding “widespread” incidents affecting multiple clients. Experts warn, however, that as AI-driven losses grow, insurers may increasingly contest claims in court. Aaron Le Marquer, head of insurance disputes at law firm Stewarts, told the Financial Times that it could take a major systemic AI failure before insurers acknowledge such events were never intended to be covered.
