Cowbell launches standalone AI agent liability policy for SMEs

The standalone coverage addresses scenarios that fall outside standard cyber and CGL policies as ISO exclusions spread
Cowbell launches standalone AI agent liability policy for SMEs
DIGITAL TRANSFORMATION
By Mark Rosanes
Nov 10, 2026

Small businesses handing control to AI agents have been running ahead of their insurance coverage. Cowbell, a California-based cyber insurer focused on small and medium-sized enterprises (SMEs), is moving to close that gap. The company is launching a standalone AI liability policy in January 2027 built for agentic AI: systems that execute tasks, move money, file documents, and interact with customers without a human approving each step.

The product, called Cowbell AI liability insurance, will be available exclusively through wholesale brokers on a non-admitted basis. It is built around a defined “AI incident” coverage trigger, combining first-party protection for a business’ own losses with liability coverage for third-party claims arising from the same event.

A gap that keeps getting wider

The timing reflects a market that has been pulling in two directions at once. ISO endorsement CG 40 47, which became available for carrier attachment in January 2026, allows insurers to remove coverage for bodily injury, property damage, and advertising injury arising from generative AI under standard commercial general liability (CGL) policies. 

A nationwide review of state filings found 2,369 generative AI exclusion records already in force across 49 states by mid-2026, with no admitted standalone product available to replace the coverage the exclusions remove. 

That exclusion activity is concentrated at the CGL level, but it has ripple effects across a client’s entire program. If a general liability policy excludes generative AI, the question becomes whether the client’s errors and omissions (E&O) or cyber policy picks up the exposure, or whether it falls between policies. For SME clients deploying AI agents in customer service, payments, or back-office operations, that question has no clean answer under most current programs. 

Cowbell’s policy is designed to sit where existing coverage does not reach. The trigger is any event in which an AI tool exceeds its authorized scope, acts without required approval, exposes data without authorization, produces unsafe outputs, or fails to operate within established controls. It also covers adversarial manipulation and model drift, where the AI is not simply malfunctioning but has been compromised or has degraded over time.

Where cyber and CGL fall short

Three illustrative scenarios from the announcement give a practical sense of the exposures the policy targets:

  • an AI accounts payable agent misinterprets approval statuses and releases duplicate payments, some of which are unrecoverable after a supplier enters liquidation
  • an AI quoting agent loops repeatedly, launching additional agents and generating unauthorized usage charges before the system is suspended
  • an AI bookkeeping agent files client tax returns using outdated rates without human review, triggering penalties and client lawsuits

Each scenario involves a loss that cuts across multiple lines. The first is a direct transaction loss. The second involves what the policy terms “runaway cost,” meaning unauthorized usage charges that accumulate when an AI loops or escalates tasks without hitting a spending limit. The policy also covers LLMjacking, where stolen credentials allow unauthorized third parties to run up API charges. The third involves third-party liability claims. 

Third-party coverage under the new policy may respond to claims alleging algorithmic discrimination, copyright or trademark infringement, bodily injury, or property damage from reliance on AI output. It may also extend to regulatory proceedings and, where permitted by law, certain fines, penalties, and punitive damages.

Three proprietary factors assess AI risk exposure

Cowbell assesses AI liability exposures using three proprietary factors evaluating AI observability, autonomy, and governance, or essentially, how a business monitors its AI tools, what authority those tools have to act independently, and what controls are in place.

This mirrors a broader underwriting shift already underway in the market. According to Ashley Bennett Jones, senior vice president for AI risk and advisory at Lockton’s cyber and technology practice, agentic AI deployments are increasingly becoming a distinguishing factor in how carriers evaluate commercial accounts, though interest in standalone AI liability products has not yet translated into accelerating take-up. 

The wholesale-only, non-admitted distribution reflects where the product sits in the market. Without admitted-market equivalents available, specialty placement is currently the only route for businesses seeking standalone AI agent coverage. That has a direct implication for placement: retail brokers without an existing wholesale relationship will need to establish one before the January 2027 launch. Coverage details are based on Cowbell’s pre-launch description. The policy is not yet live and final terms will govern at issuance.

Related Stories
Free newsletter

We'll keep you up-to-date with the latest breaking news, cutting edge opinion, and expert analysis affecting both your business and the industry as whole.

Free newsletter

Our daily newsletter is FREE and keeps you up - to - date with the world of Insurance. Please complete the form below and click on subscribe for daily newsletters from IB US.