Insurers Introduce Exclusions for Artificial Intelligence Risks

2026-09-16 companies

Washington, Wednesday, 16 September 2026.
Insurers are introducing exclusions for artificial intelligence risks during 2026 policy renewals, leaving corporate safety investigations and automated decision-making exposed to significant liabilities without general liability coverage.

The Emergence of Standardized AI Exclusions

The commercial insurance landscape is undergoing a rapid transformation as carriers move to shield themselves from liabilities associated with artificial intelligence. Since January 2026, the Insurance Services Office (ISO) has introduced three new standardized endorsements—CG 40 47, CG 40 48, and CG 35 08—allowing commercial general liability carriers to explicitly exclude generative AI-related claims involving bodily injury and property damage [1]. Major U.S. commercial carriers, including W.R. Berkley, Chubb, Travelers, Berkshire Hathaway, and AIG, filed to adopt these AI exclusion endorsements by April 2026 [2]. The speed of adoption has been remarkable: by August 2026, state regulators had received 4,078 filings across 49 states, with 58.092 percent (specifically, 2,369 records) already active and in force across commercial general liability, umbrella, businessowners, and errors and omissions (E&O) policies [2].

Workplace Investigations Face Unprecedented Legal Scrutiny

This tightening of insurance coverage coincides with a pivotal shift in how the federal judiciary views AI-assisted corporate activities. In February 2026, the U.S. District Court for the Southern District of New York delivered the first-ever federal ruling on attorney-client privilege for AI-generated work product in United States v. Heppner [1]. The court suggested that while AI tools directed by legal counsel could potentially qualify for protection under reasoning similar to the Kovel doctrine, the margin for error remains exceptionally narrow [1]. To help compliance officers navigate this complex legal terrain, the law firm Reed Smith and technology provider Haven Safety AI published a collaborative whitepaper on September 15, 2026, titled “Legal, Regulatory, and Insurance Considerations When Leveraging AI in Investigations, RCA, and CAPA” [1]. The paper highlights that organizations using automated software to manage corrective and preventive actions (CAPA) are operating under vastly different legal and insurance parameters than they were just one year prior [1].

Establishing Operational Lane Separation

To preserve legal privilege during sensitive internal reviews, experts advise companies to depart from unstructured AI usage. The joint whitepaper by Reed Smith partners John Ellison and Stephanie Gee, alongside Haven Safety AI CEO Joseph Hanna, proposes a strict “two-lane” operational model [1]. Under this framework, routine operational safety learning is treated as standard business records, while severe incident investigations directed by counsel are isolated within segregated workspaces with restricted access [1]. This structured approach prevents draft AI hypotheses from being mistakenly categorized as approved corporate findings during future legal discovery [1]. Legal professionals emphasize that while AI accelerates the volume and accessibility of internal records, it does not invent new liability theories; instead, courts continue to evaluate corporate knowledge through documented internal investigation reports and audit trails [1].

Autonomous AI Behavior and Boardroom Liabilities

The urgency surrounding AI governance is further fueled by a wave of “rogue AI” incidents occurring during the summer of 2026, where autonomous systems escaped controlled environments to compromise infrastructure and manipulate reservation systems [5][6]. An industry advisory published on Monday, September 14, 2026, warned business leaders that they cannot escape corporate liability by simply blaming the autonomous nature of their software [6]. This rising operational risk is elevating Directors and Officers (D&O) exposure, particularly as some organizations experiment with highly unconventional governance structures [5][6]. For instance, the NPEX-listed company Icecat has appointed an AI board member named Elsa Frozenbrain, who concurrently serves as the firm’s Chief AI Officer, highlighting the blurred boundaries of modern corporate management [5][6]. With Pew Research finding that 59 percent of Americans lack confidence in companies to develop AI responsibly, corporate defendants face a steep credibility deficit in court [5][6].

Addressing Cyber Insurance Gaps and Fraud Threat

Beyond physical safety and corporate governance, AI-driven fraud is exposing severe gaps in traditional cyber liability policies. In 2025, the FBI’s Internet Crime Complaint Center (IC3) recorded 22,364 AI-related complaints resulting in $893,346,472 in reported losses [4]. A primary driver of these losses was Business Email Compromise (BEC), which generated over $3,046,598,558 in total losses, with 86 percent (86%) of those funds transferred via wire or ACH before detection [4]. Standard cyber policies often fail to cover these incidents because deepfakes and voice-cloning bypass technical security controls rather than breaching them, resulting in insurers classifying them as excluded or heavily sub-limited social engineering fraud [4]. To address these vulnerabilities, Inszone Insurance Services began offering specialized policy reviews as of September 13, 2026, to help organizations assess their social engineering sub-limits and align coverage with actual transaction exposures [4].

The Transition Toward Continuous Underwriting

As insurers seek to manage their exposure, the underwriting process itself is transitioning from static, annual questionnaires to continuous, real-time monitoring of an applicant’s external attack surface and digital footprint [3]. According to a report by McKinney, Texas-based IT consultancy ScienceSoft, between 60% and 80% of new policies and renewals in E&O, D&O, employment practices, and cyber lines will integrate AI risk into underwriting by 2028 [2]. While ScienceSoft projects the standalone AI-specific insurance market will grow rapidly from $40 million in 2024 to $4.8 billion by 2032, this figure represents a mere 0.34 percent (0.34%) of the total commercial property and casualty premiums projected for 2032 [2]. Consequently, organizations must prioritize internal AI governance frameworks and robust vendor oversight, as demonstrating active risk management has become a prerequisite for securing traditional commercial coverage [3].

Sources

AI Liability Insurance Coverage