AI agent liability insurance: Cowbell's policy explained

On November 10, the cyber insurer Cowbell announced a standalone policy for small and midsized US businesses whose AI agents do something expensive on their own. It’s called Cowbell AI Liability Insurance, and according to the company’s press release it goes on sale on January 1, 2027.

That’s the headline. The details matter more, so here they are as the questions a business owner would ask.

What counts as a covered event?

Cowbell calls the trigger an “AI Incident.” Per the release, that means an AI tool exceeding its authorized scope, acting without required approval, exposing data, producing inaccurate or unsafe output, or failing to stay inside its controls. Adversarial manipulation and model drift are on the list too.

Cowbell gave three examples. An accounts-payable agent misreads approval statuses and pays suppliers twice, and the supplier then goes into liquidation. A quoting agent loops, spawns more agents and sends customers wrong quotes while usage charges pile up. A bookkeeping agent files client tax returns with outdated rates and the clients sue.

What does it pay for?

On the first-party side, the policy covers investigation and containment, correcting or withdrawing bad output, unrecovered funds from unapproved or manipulated actions, and two items that are very specific to AI. One is “runaway cost,” meaning usage charges from looping tasks or spending limits that didn’t hold. The other is LLMjacking, where someone uses stolen credentials or API keys to run AI workloads on your bill.

Third-party coverage may pay defense costs and settlements for claims over algorithmic discrimination, copyright or trademark infringement, bodily injury or property damage. Where the law allows, SiliconAngle reported, it can also reach regulatory fines and penalties.

Who can buy it, and how?

Only through wholesale brokers, and on a non-admitted basis, according to Cowbell. Insurance Business pointed out what that means in practice: retail brokers without a wholesale relationship will need one before January, and specialty placement is currently the only route to standalone agent coverage.

Cowbell also says underwriting will use three scores it calls Cowbell Factors, covering observability, autonomy and governance. Roughly, how closely you watch your agents, how much they’re allowed to do, and what rules sit around them. Businesses can see their scores by connecting Cowbell to tools like Gemini, Claude and ChatGPT.

What’s still missing?

Question a buyer would ask	What the sources say
Premium	Not stated in the release or the coverage I read
Policy limits and retentions	Not stated; “subject to the terms, conditions, limits, retentions, and exclusions”
Exclusions	Not stated
Availability	January 1, 2027, wholesale brokers only

Insurance Business added that the coverage details come from Cowbell’s pre-launch description, and that final terms will govern at issuance. It’s a brochure. Not a contract.

Why now?

Insurance Business wrote that an ISO endorsement known as CG 40 47, available to carriers since January 2026, lets insurers strip bodily injury, property damage and advertising injury arising from generative AI out of standard commercial general liability policies. If your regular policy has that endorsement, an agent’s mistake may not be covered there at all. It’s worth asking your broker which version you hold.

My read: the product matters less than the questions behind it. An insurer asking how much an agent can do without a human, and whether you can see what it did, is handing you a better checklist than most vendor safety pages do. Will the premium make sense for a ten-person shop? Nobody outside Cowbell can say yet.

The examples in the release are also dull, which I think is the point. They’re ordinary software failures with an agent attached, the same kind of problem as an OpenAI agent reaching Australia’s Medicare portal, where OpenAI said it found no evidence patient records were accessed. Agents that touch money and customer records are now an insurable risk. The first policy shows up before anyone has published loss data.

Cowbell will insure small businesses against their own AI agents. Prices and limits aren't public yet.