AI is too risky to insure, say people whose job is insuring risk (techcrunch.com)

Major insurers including AIG, Great American and WR Berkley have told U.S. regulators they want permission to exclude AI-related liabilities from corporate insurance policies, arguing that modern AI models are “too much of a black box” to underwrite. The move follows high-profile mishaps—Google’s AI Overview allegedly made false accusations that spawned a $110M lawsuit, Air Canada honored a chatbot-invented discount, and fraudsters used a deepfake executive to steal $25M from engineering firm Arup—illustrating liabilities from hallucinations, erroneous automation and convincingly realistic synthetic media. The core concern is systemic, not one-off losses: insurers fear a widely deployed model fault could trigger thousands of simultaneous claims, a correlated risk they can’t absorb even if they can handle large single losses. Practically, exclusions would shift liability to vendors, leave firms exposed or uninsurable, and likely raise premium costs while accelerating demand for technical risk controls: model explainability, robust testing, provenance and audit trails, contractual indemnities, and regulatory standards. The industry signal could force rapid changes in procurement, governance and AI safety engineering as insurers push for measurable mitigations before they’ll cover AI-driven operations.