The short answer
An AI startup's program starts with technology errors and omissions, usually called tech E&O, at $1M to $5M per claim, because a model's output causing a customer a financial loss is the live exposure. Cyber liability sits beside it at $1M or more once you hold customer data in production. General liability at $1M per occurrence and $2M aggregate arrives with the first lease or contract that demands a certificate, and directors and officers coverage at $1M to $3M arrives with the first priced round. Every one of those forms has to be read for an AI exclusion before it binds.
When your product is the work
A traditional software company ships tools and the customer does the work. An AI company ships the work itself: an agent answers the support ticket and a model approves the transaction. When that work goes wrong, the customer's loss traces directly to your product's decision, and the claim comes to you.
The contracts make this sharper. Model providers' terms of service push liability downstream to whoever deployed the model. Your customer's agreement is with you, and the lab that trained the underlying model is not a party to it. Between those two documents, the company in the middle holds the exposure.
None of this means AI companies are uninsurable. The lines that respond are the same ones technology companies have carried for decades. What changes is how much the policy wording matters, because forms written before agents existed are now being asked questions they were never designed to answer.
The four lines, in the order they matter
| Tech E&O | The anchor. It is the line built to respond when your technology's work causes a customer a financial loss, and a model's output is your technology's work. For an AI company this is the policy that faces the live exposure. | $1M to $5M per claim; triggered by the first customer contract |
|---|---|---|
| Cyber liability | Often sold with tech E&O as one combined form, and it answers a different claim: your systems breached, customer data exposed, ransomware, and the response costs that follow. An AI product that touches customer data in production needs both parts, and a combined form needs both parts read. | $1M or more; triggered by customer data in production |
| General liability | Covers bodily injury and damage to other people's property. Software rarely triggers it, but nearly every office lease and customer contract requires a certificate of it, so it arrives early for practical reasons rather than exposure reasons. | $1M per occurrence and $2M aggregate; triggered by the first lease |
| Directors and officers | Protects company leadership when they are sued personally over decisions running the business. For AI companies it carries extra weight: public claims about what your AI can do are statements to investors, and overstated capability claims are an active area for suits and regulator attention. | $1M to $3M; term sheets require it at the first priced round |
What enterprise procurement will ask for
The first time insurance blocks a deal is usually an enterprise security and vendor review. The common asks are tech E&O, cyber, and general liability certificates at stated limits, often between one and five million dollars depending on the contract. Some agreements name additional insureds or require specific endorsements.
Expect the security questionnaire to ask about the model too: what it does with customer data, whether its outputs are reviewed, how it is monitored in production. The same reviewer reads the insurance exhibit and the questionnaire. When the two describe different products, the deal stalls until they agree.
Have the program placed before the first enterprise deal reaches procurement, so insurance stays off the critical path of a signature. A certificate from a program already in place is a short broker task. A new line requested mid-negotiation needs underwriting first, and procurement waits for it.
What the program costs
A quote is priced on the company in front of the underwriter, one policy at a time. Tech E&O is the policy an AI product moves most, because the underwriter is pricing a product that makes decisions for the customer under contracts that demand $1M to $5M in limits. Cyber follows revenue and the customer data you hold, D&O follows the money you have raised, and general liability follows where your people work.
Three things you control move the price. A higher retention lowers it, because you take the first part of each claim yourself. Clear answers lower it, because an underwriter who cannot see how the model is tested and who reviews its output prices the worst case. Buying limits when a contract asks for them keeps you from paying for cover nobody has requested.
The drivers are revenue, what the product does inside the customer's operation, how much customer data it touches, the limits the contracts demand, and whether the form covers AI affirmatively or silently. A form with an AI exclusion can be the cheapest quote on the table. It is also the one that pays nothing on the claim you bought it for.
What sets the price, by policy
| Tech E&O | Revenue, the contracts you sign, and what the model decides for a customer | The policy an AI product moves most |
|---|---|---|
| Cyber | Revenue, the customer data you hold, and your security controls | |
| D&O | Capital raised, the board, and financial position | |
| General liability | Where your people work and what you sell | |
| Workers' compensation | Payroll and the job each person does |
AI exclusions, absolute and limited
Standard technology forms predate all of this, and carriers quote them anyway. Some carriers have begun filing AI-specific exclusions, and nothing flags them at purchase: an exclusion only takes effect when there is a loss, and then it decides whether the claim is paid. The filings are no longer scattered. The standard-forms bureau's generative AI exclusions for general liability took effect in January 2026 (Independent Agent magazine, 2025-10-21), and more than 60 insurance groups had filed to adopt them by July 2026 (The Insurer, 2026-07-23). A global reinsurer's research note (2026-08-26) counted more than 80% of those filings approved, with 41 groups adopting and 20 delaying.
Two shapes of exclusion are in circulation, and they behave very differently at claim time. A limited or generative AI exclusion removes claims arising from content a model created, and it usually leaves the rest of the policy alone. An absolute AI exclusion removes any claim that arises from, relates to, or involves artificial intelligence in any way, and one such exclusion on management-liability forms was reported by Hunton Andrews Kurth on 2025-05-28. For a company whose product is a model, an absolute exclusion on D&O or E&O empties the policy. The word to look for is absolute.
Two quotes with the same premium and the same limits can behave differently at claim time, depending on how each form defines professional services and whether either mentions machine learning at all. Every form should be read before anything binds, by someone who knows what an AI exclusion looks like.
Agents raise the stakes again. An agent that writes to a customer's CRM or moves money produces a wrong action, and nobody reviews it before the loss. The standard-forms bureau is now studying exclusions aimed at agentic AI (The Insurer, 2026-07-10). If your product acts inside customer systems, our guide to insurance for AI agents covers what tech E&O does and does not do for it.
When to buy each line
Pre-revenue, documents drive the purchase: a lease wants general liability, and an investor wants D&O. The first customer contract is the natural trigger for tech E&O and cyber, because that is when your product's failure starts having someone to owe.
The placement itself is a description problem. Tell the underwriter what the model does, what it decides without a person, what data it sees, and how it is evaluated before and after release. We place through insurers who cover AI on purpose, read the professional services definition and every AI endorsement ourselves, and send you the exclusion language before you bind.
This page describes coverage in general terms. It is not an offer of insurance, and carrier appetite, policy wording, licensing, and availability govern every quote and every claim.
Common questions
What insurance does an AI startup need first?
Tech E&O and cyber anchor the program, because the live exposures are your model's output causing a customer a loss and your systems exposing data. General liability arrives with the first lease or contract that demands a certificate, and D&O arrives with the first priced round.
How much does insurance cost for an AI startup?
It is priced one policy at a time, on your revenue, what the product decides, how much customer data it touches, and the limits your contracts demand. An AI company with enterprise contracts pays more than an ordinary software startup of the same size, mostly through tech E&O. The quote for your company comes from the carriers' underwriters.
Can an AI company get insurance at all?
Yes. The lines are standard. The placement work is presenting the product accurately and reading the forms that come back, because insurer appetite for AI risk varies widely and exclusions now appear in many filings.
What is the difference between an absolute AI exclusion and a generative AI exclusion?
A generative or limited AI exclusion removes claims arising from content a model created and usually leaves the rest of the form intact. An absolute exclusion removes any claim that arises from or relates to artificial intelligence in any way, which for a company whose product is a model can empty the policy. The first is negotiable at the edges; the second usually means finding a different insurer.
Does general liability cover mistakes our model makes?
No. General liability responds to bodily injury and damage to other people's property, and the standard-forms bureau's generative AI exclusions for that line took effect in January 2026 (Independent Agent magazine, 2025-10-21). A customer's financial loss from a wrong output is a tech E&O claim, subject to that form's wording.
Do we need cyber insurance if the model provider hosts the model?
Yes. The provider's terms leave your prompts, your customer data, and your integration on your side of the line. When a prompt-injection attack turns your product into the leak, the notification and the claim come to you, and cyber is the line that responds, subject to how the policy defines a security event.
Terms in this guide
Sources
- 01Independent Agent magazine, on the standard-forms bureau's generative AI exclusions for general liability, 2025-10-21
- 02The Insurer (subscription), "More than 60 P&C insurance groups file to adopt AI exclusions", 2026-07-23
- 03A global reinsurer's research note on approval rates and adoption of AI exclusions, 2026-08-26
- 04Hunton Andrews Kurth, insurance recovery blog on the proliferation of AI exclusions, including an absolute AI exclusion on management-liability forms, 2025-05-28
- 05The Insurer, on the standard-forms bureau weighing exclusions for agentic AI, 2026-07-10
Read next
- Insurance for AI agentsHow to insure an AI agent that takes actions inside customer systems: which lines respond to a wrong action at scale, what tech E&O does and does not do for agents, and what to document before you apply.
- Your renewal added an AI exclusion. Now what?How to read a generative or absolute AI exclusion on a renewal, the steps to take before the renewal date, what a buy-back costs, and where affirmative AI coverage comes from.