AI risks make some insurers wary of corporate liability

RAND wants better data to price machine-made mishaps, apocalypse not included Thomas Claburn Thomas Claburn AI AND SOFTWARE REPORTER Published fri // UTC German firm files for insolvency, blames cybercrims who shut down production for 6 weeks July 13, 2026 Estonia intends to recognize AI agents with digital IDs June 18, 2026 Google found liable for bad AI Overview results. And if AI ends up being uninsurable, the AI industry will have to moderate its ambitions and sales targets while corporate customers delay AI projects to fulfill their fiduciary obligations. RAND says its report was motivated , particularly in the US. Apocalyptic scenarios aside, the report says that companies are already grappling with AI-related incidents and related litigation while insurers, regulators, brokers, and policyholders puzzle over risks, coverage, and rules. Berkley has already introduced exclusions in its D&O (directors and officers), E&O (errors and omissions), and Fiduciary Liability insurance products to exclude coverage for "any actual or alleged use, deployment, or development of Artificial Intelligence." During the company's Q4 2025 earnings call, CEO W. Robert Berkley highlighted the need for underwriters to understand "the impact that [new technologies like AI are] having on our insureds, what it means for risk, and our ability to fully understand that risk so we can control it, select it, and price for it." The RAND report observes that some insurance carriers are excluding AI-related harms, noting that in January 2026, Verisk/ISO – whose standardized forms appear in more than 80 percent of US property and casualty policies – introduced optional language carriers can adopt to exclude bodily injury, property damage, and other harms arising from generative AI. Not all insurers are deciding not to cover AI harms. As RAND points out, coverage gaps are being filled a handle on the risk calculations. To illustrate those risks, the report points to the Artificial Intelligence Incident Database (AIIDB) which lists tral-world harm or near-harm caused . At the time of writing, the AIIDB lists 713 incidents drawn from more than 6,000 reports and covers AI use beyond chatbots. Among these generative AI events, the categorical breakdown includes: misinformation and manipulation (586), deepfakes and synthetic media (346), deepfake enabled misinformation (333), hallucination and factual error (215), harmful content (92), agentic and autonomous failure (84), privacy and data leak (58), bias and discrimination (47), copyright and IP (20), other and unclassified (14), and wrongful AI attribution (7). The total of those events exceeds 713 because some incidents span multiple categories. Then there are the 250 or so US lawsuits related to AI, largely related to copyright and IP, but also touching on privacy/surveillance, fraud and deception, negligence and product liability, discrimination and civil rights, and contracts/trade secrets.