Insurance for companies whose product makes decisions.
AI companies are being underwritten by a market that is still writing its own rules. Some carriers exclude AI outright. Some cover it affirmatively. Most are silent. Knowing which one you're holding is the entire job.
3
Ways carriers treat AI: exclude, stay silent, or cover it
48h
Typical time to quote a prepared AI submission
$2M
Most common enterprise E&O requirement at Series A
Broker response
Same business day
Answered by a broker, in writing.
Quote turnaround
3–10 business days
From a complete submission to terms in hand.
Certificates
Same day
Standard requests issued without an email chain.
Renewal process
Begins 90 days out
Not the week before your policy expires.
What actually goes wrong at companies like yours
Not a generic risk list. These are the exposures underwriters ask about and the ones that produce claims.
Model output liability
When a customer relies on an inference, a recommendation, or generated content and suffers financial loss, the claim lands on your Tech E&O — if the form doesn't exclude AI.
Silent AI and express AI exclusions
Carriers are actively adding artificial intelligence exclusions to E&O and cyber forms. Two quotes at the same limit and premium can behave in opposite ways at claim time.
Training data and IP provenance
Copyright claims tied to training corpora and generated outputs are live litigation. Most E&O forms cover unintentional copyright infringement; almost none cover patent, and several now carve back generative output.
Bias and discriminatory outcomes
Models used in hiring, lending, insurance, housing, or healthcare create discrimination exposure that spans E&O, EPLI, and regulatory coverage — often falling between all three.
Regulatory exposure
The EU AI Act, Colorado's AI Act, and a growing set of state algorithmic accountability laws create investigation and penalty exposure that many cyber forms address only for privacy events.
Customer indemnity obligations
Enterprise buyers increasingly demand IP indemnification for model output. Your standard terms need to line up with what your policy will actually fund.
Data pipeline and vendor concentration
Dependence on a small number of model providers and cloud regions creates business interruption exposure that only dependent BI coverage addresses.
Compute and infrastructure spend
Large committed compute contracts make business interruption and contingent BI limits materially more important than headcount alone would suggest.
What we typically place
A starting structure for ai companies. Your contracts, data footprint and headcount move it — which is exactly what the submission is for.
Build your programmeThe AI exclusion problem
Over the last two renewal cycles, carriers have moved in three different directions on artificial intelligence. Some added broad exclusions removing any claim arising from the development, deployment, or use of AI. Some stayed silent, leaving the question to be resolved by a coverage lawyer after a claim. A smaller group began offering affirmative grants that name AI and machine learning explicitly.
This produces a market where price is a poor signal of protection. We have seen two quotes for the same company, at the same limit, within a few percent on premium, where one form would respond to a model-output claim and the other would not. For a company whose entire product is model output, that is not a detail.
Our standard practice is to pull the AI-relevant language from the quotes we present, put it side by side, and tell you in plain terms which of the three you're being offered. If a carrier won't offer affirmative language, that is a reason to keep shopping — not a reason to accept the exclusion quietly.
What underwriters ask AI companies
Submissions for AI companies get more scrutiny than a comparable SaaS business. Underwriters want to know whether outputs are advisory or automated, whether a human reviews consequential decisions, what the training data provenance is, whether you fine-tune or serve third-party models, and how you handle customer data used in inference.
They also want to see the contract stack: your standard terms, your limitation of liability, your IP indemnity, and any commitments you've made about model accuracy. A well-prepared submission that answers these questions before they're asked prices better and binds faster. Weak submissions get referred, and referred submissions get expensive.
We build the narrative for you. You answer a structured intake once, and we turn it into the technical description, the AI supplement, and the control narrative that carriers want to see.
Where enterprise procurement is heading
Enterprise buyers have started adding AI-specific language to vendor agreements: accuracy representations, human-oversight commitments, model-change notification, training-data warranties, and indemnity for third-party IP claims arising from output. Some now require evidence of insurance that does not exclude AI.
That last one matters. A certificate of insurance says nothing about exclusions. When a customer's security or legal team asks whether your E&O covers AI, the answer needs to be documented — and increasingly, they will ask for the endorsement, not the certificate.
What your customers will ask you to carry
The limits that show up most often in insurance exhibits for ai companies. Send us the exhibit before you sign and we'll tell you whether you comply, what compliance costs, and what is worth negotiating.
| Requirement | Typical ask |
|---|---|
| Technology E&O | $2M – $5M, often with express AI coverage |
| Cyber liability | $2M – $5M, combined or separate |
| General liability | $1M per occurrence / $2M aggregate |
| Workers' compensation | Statutory + $1M employers liability |
| Umbrella / excess | $1M – $5M over GL, auto, and EL |
| Crime / fidelity | $500k – $1M where funds are handled |
AI companies, answered
Anything else, send it to a broker and get a written answer within the business day.
It depends entirely on the form. Some carriers now exclude AI explicitly. Others are silent, which means the answer gets decided during a claim. A growing set offers affirmative coverage. We read the language on the quotes we bring you and tell you which one you have.
Partially. Most E&O forms cover unintentional copyright infringement arising from your services or product. Several markets now carve back generative output specifically. Patent is excluded on essentially every form. This is a language exercise, not a limit exercise.
Quite a lot. Discrimination exposure spans professional liability, employment practices, and regulatory coverage, and the trigger depends on who is bringing the claim — your customer, their applicant, or a regulator. We map the exposure explicitly rather than assuming one policy catches it.
D&O once you take institutional money, workers' comp once you have employees, and cyber once you hold anyone else's data. E&O usually waits until you have paying customers or a signed contract that requires it.
We treat it the way we treat GDPR: a regulatory exposure that needs to be reflected in the cyber and E&O regulatory coverage grants, and disclosed accurately in the submission. We'll tell you what your policies actually respond to and where the gap is.
Get a programme built for ai companies.
Tell us about the business once. A broker reviews it the same day and comes back with a plan, a timeline and what we need to take it to market.