Cowbell to launch standalone AI agent liability policy for small businesses
The California-based insurer is addressing a growing coverage gap for companies using autonomous AI systems that move money or file documents without human oversight.
The short version
- The policy targets 'agentic AI' systems that execute tasks like payments and customer interactions without human approval for every step.
- Coverage triggers on events like unauthorized data exposure, 'runaway costs' from infinite loops, and adversarial manipulation.
- The product will be available exclusively through wholesale brokers on a non-admitted basis starting in January 2027.
- The policy covers both first-party losses for the business and third-party liability claims like copyright infringement or discrimination.
Why it matters
As businesses move from simple chatbots to autonomous agents, standard insurance policies are actively removing AI coverage. This new product provides a specific safety net for the financial and legal risks of letting AI operate independently.
Closing the insurance gap
Small and medium-sized enterprises (SMEs) are increasingly handing over operational tasks to AI agents. These are systems capable of moving money, filing documents, and interacting with customers without a human approving every individual action. However, many of these businesses are currently operating ahead of their existing insurance coverage.
Cowbell, a California-based cyber insurer, plans to address this by launching a standalone AI liability policy in January 2027. The company aims to provide a specific safety net for the unique risks posed by agentic AI. This move comes as the broader insurance market begins to distance itself from AI-related risks in standard packages.
As Insurance Business reported, the timing is driven by a shift in how traditional policies are written. Insurers are increasingly using specific endorsements to remove coverage for injuries or damages caused by generative AI. This leaves a vacuum where businesses may find themselves with no clear protection when an autonomous agent causes a loss.
Defining an AI incident
The Cowbell policy is built around a specific definition of an "AI incident." This trigger covers any event where an AI tool exceeds its authorized scope or acts without the required human approval. It also applies if the tool exposes data without authorization or produces unsafe outputs that violate established controls.
The policy is designed to handle complex technical failures like model drift. Model drift occurs when an AI's performance degrades over time, meaning the system is no longer operating as intended. It also covers adversarial manipulation, which happens when an external party compromises the AI to perform unauthorized actions.
By defining these specific triggers, the policy aims to sit where current programs fall short. Many SMEs find that if their general liability policy excludes AI, it is unclear if their cyber or errors and omissions policies will pick up the slack. Cowbell intends to provide a clear answer to that question by offering a dedicated product.
Practical risk scenarios
To illustrate the types of risks covered, Cowbell highlighted three specific scenarios. The first involves direct transaction losses where an agent might move funds incorrectly. The second involves "runaway costs," which are unauthorized usage charges that accumulate when an AI enters an infinite loop or escalates tasks without hitting a spending limit.
The policy also addresses "LLMjacking." This occurs when a bad actor steals credentials to run up high API charges on a company's account. By covering these specific technical and financial failures, the policy protects the business from the immediate costs of a malfunctioning or hijacked system.
Beyond internal costs, the policy covers third-party liability claims. This includes allegations of algorithmic discrimination, copyright infringement, or trademark violations resulting from AI output. It may even extend to regulatory proceedings and certain fines or penalties, depending on what is permitted by law.
Underwriting and distribution
Cowbell does not treat all AI deployments the same way. The company assesses risk using three proprietary factors: AI observability, autonomy, and governance. These factors essentially measure how a business monitors its tools, how much independent authority those tools have, and what specific controls are in place to prevent errors.
The distribution of this policy is notable because it will be available exclusively through wholesale brokers on a non-admitted basis. This means that retail brokers who do not already have a wholesale relationship will need to establish one before the January 2027 launch. This specialty placement is currently the only route for businesses seeking this specific type of coverage.
While interest in standalone AI products is growing, it has not yet translated into a massive surge in adoption. Industry experts note that agentic AI deployments are becoming a key factor in how carriers evaluate accounts. For now, the product serves as a specialized tool for businesses ready to move beyond basic chatbots into fully autonomous workflows.
What to watch
- Whether more insurers will follow Cowbell's lead in creating admitted standalone AI products.
- How the definition of 'runaway costs' evolves as API pricing models change.
- The actual take-up rate of these policies among SMEs as agentic AI becomes more common.
Try it
Check your current commercial general liability or cyber insurance policy to see if it contains any specific exclusions for generative AI or autonomous systems.
Sources and links
First reported by Insurance Business. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.