Insurers balk at paying out huge settlements for claims against AI firms (arstechnica.com)

OpenAI and Anthropic are preparing to rely on investor cash to cover potential multibillion-dollar legal claims after insurers signaled they won’t underwrite the full scale of AI-related risk. While both startups hold traditional policies—OpenAI reportedly secured up to $300 million of emerging AI-risk cover per some sources (a figure others dispute)—insurance professionals say that amount is far short of what would be needed for “nuclear verdict” style damages. Brokers like Aon acknowledge the market lacks capacity for model providers and worry most about systemic, correlated, aggregated losses that could arise if widely deployed models make harmful mistakes. For the AI/ML community this is material: it exposes a gap between technical risk and financial risk transfer. Insurers’ reticence raises the prospect that founders and investors, not policies, will backstop catastrophic liability, changing startup economics and investor due diligence. It also creates pressure for model providers to harden engineering and governance — better testing, red-teaming, interpretability and incident containment — to reduce correlation of failure modes and make risks insurable. Longer term, expect new insurance products, higher premiums, and tighter contractual risk-sharing (and possibly regulation) as the industry grapples with quantifying and limiting large-scale harms from deployed AI systems.