AI death claims hit 2,000

Gartner research projects that claims for deaths involving artificial intelligence will exceed 2,000 globally by the end of 2026. Alissa Lugo, a senior director analyst in Gartner's legal and compliance practice, defines these cases as instances where AI played some role in a fatality. The number is expected to grow as the technology becomes more embedded in daily operations.

Anthropic, the AI company that filed for an initial public offering in late October, addressed this uncertainty in its S-1 filing. The document states that models may behave in unintended ways, resulting in substantial liability. Anthropic warned that its insurance policies might not provide sufficient coverage to mitigate potential liabilities, particularly where a single incident causes disproportionate harm. The company noted that the insurance market for AI-specific risks is nascent, and it may be unable to obtain coverage on commercially reasonable terms.

Industry experts compare the current AI liability situation to the early days of cyber insurance. Dennis Hendrickson of Hendrickson Insurance noted that cyber policies currently cover AI losses only in specific contexts like data breaches or wire fraud. He described the lack of coverage for bodily injury as a gaping hole, as cyber insurance was built for data and financial losses rather than physical harm.

Kara Higginbotham, head of professional liability at Zurich North America, said insurers are handling AI exposures differently across various product lines. She explained that a gap exists because carriers are wary of writing bodily harm into policies for scenarios that are too new to underwrite effectively. General liability policies could potentially cover these incidents, but insurers often require proof of baseline controls and risk management practices before extending coverage.

Vikram Desai, global cyber strategy lead at Accenture, observed that the focus has shifted from integrating AI into business to managing what happens when it fails. He stated that AI is following a compressed version of the cyber insurance curve, where insurers initially excluded the risk and then sold it back as a separate product. Underwriters are now asking companies where AI is used and who oversees it to determine risk levels.

The industry remains divided on whether AI insurance will become a standalone product or fold into existing cyber policies. Jaymin Kim from Marsh's cyber practice argued that AI is a technology through which familiar harms manifest, making it difficult to isolate in the causal chain of loss. She suggested that creating separate policies for every AI-touching process could become a nightmare for customers managing competing coverage.

Hendrickson predicted that insurers will set the standards for AI use before government regulations do. He expects a deluge of claims that large carriers are not currently prepared to handle. For now, independent agents are dealing with the immediate fallout as companies navigate unclear coverage boundaries and emerging exclusions.