Insurance for AI Companies
AI companies sell outputs nobody wrote by hand, built on data and models that often belong to someone else. We help AI companies compare tech E&O, cyber, media and D&O options for wrong outputs, data and IP disputes, and the investors behind the company.
AI businesses we work with
Most AI companies in California do not build a foundation model from scratch. They build a product on top of one. The insurance questions change with where you sit:
- companies building apps, copilots and agents on top of a third-party model;
- teams that fine-tune or build their own models and sell access to them;
- AI consultants and integrators who deploy models inside a client’s business;
- data labeling, annotation and evaluation firms;
- robotics and hardware companies that put a model into a physical product.
A customer who is harmed by your product looks to you, whoever made the model underneath. Read your model provider’s terms closely, because they usually limit what the provider will answer for, which leaves the rest with you.
When the model gets it wrong
The most common claim against an AI company is a customer saying your output cost them money. A support agent quotes a refund policy that does not exist. An extraction tool misreads a batch of invoices and payments go out wrong. A forecasting model points a buyer at the wrong inventory. None of that is bodily injury or property damage, so general liability leaves it alone.
That claim belongs to technology errors and omissions insurance, which in IRMI’s definition reaches the services and products a tech firm sells, its media content and network security failures. For an AI company, ask that the policy’s description of your services names your AI and machine learning products, so an insurer cannot argue that a model’s output falls outside the work it agreed to cover. Our IT and technology companies page covers the rest of tech E&O.
The new generative AI exclusions
Standard policies are starting to say what they do with AI. ISO, which writes the forms many insurers build on, has published three generative AI endorsements. CG 40 47 removes bodily injury, property damage and personal and advertising injury arising out of generative AI from a general liability policy. CG 40 48 removes only personal and advertising injury, and CG 35 08 does the same work on the products and completed operations side. Insurers are filing with state regulators to use them.
For an AI company, an exclusion like that on your general liability policy can remove the coverage you expected for a defamation or advertising claim tied to your product. Read every renewal for new AI wording, and look for policies that cover AI on purpose instead of by silence. We check the endorsements on each quote before you bind.
IP and the data behind the model
Copyright owners have sued AI companies over the data used to build models and over outputs that resemble their work. Media liability, often part of a tech E&O policy, can answer for copyright, trademark and defamation claims tied to your content, while patent disputes usually sit outside it. Enterprise customers also ask for an IP indemnity in the contract, which promises that you will defend them if your product’s output infringes. Check that your insurance stands behind the promise before you sign it.
California adds a disclosure rule. Civil Code section 3111 requires the developer of a generative AI system made available to Californians to post documentation about the datasets used to build it on its website, starting January 1, 2026 and with each substantial modification. The law’s definition of a developer reaches companies that substantially modify a system, including through fine tuning. That posted summary becomes evidence if a dispute over data rights reaches court, so keep it accurate and keep records of where your data came from and what permission you have to use it.
Privacy and the data customers send you
Prompts, uploaded files and conversation logs carry personal information, contracts and trade secrets. A leaked API key, a misconfigured vector database or a prompt injection that pulls one customer’s data into another customer’s session are all breaches. They bring notice costs, forensic work and claims from the people whose data leaked.
Cyber insurance answers for those first-party costs and third-party claims. Expect underwriters to ask how long you keep prompts and logs, whether customer data feeds back into your models, how you separate one customer’s data from another’s, and who can reach production systems. Our cyber readiness checklist gathers the usual security questions.
Investors, boards and what you say about AI
Venture-backed AI companies carry a board, a cap table and investor expectations. Directors and officers insurance protects the founders and directors when investors, regulators or acquirers claim they mismanaged the company. One risk is specific to AI: claims that a company overstated what its AI can do. Keep public statements, pitch decks and sales materials in line with what the product actually does.
The largest model builders also answer to California’s Transparency in Frontier Artificial Intelligence Act (SB 53), which took effect January 1, 2026. It applies to developers of frontier models built with more than 10^26 computing operations, with extra duties for those with more than $500 million in annual revenue. Most startups fall far below both lines, but enterprise customers sometimes ask about it in security reviews. As headcount grows, employment practices liability and workers’ compensation join the package.
What to have ready for a quote
An AI application moves faster with these in hand:
- what the product does, who uses it and whose models it runs on;
- how outputs reach customers, and whether a person reviews them before they are relied on;
- a sample customer contract, with its liability cap, indemnities and insurance section;
- where your data comes from and the permissions you hold for it;
- security controls, data retention and how customer data is kept separate;
- revenue, funding stage, board makeup, and any demand letters or disputes so far.
With that list we can compare tech E&O, cyber, media and D&O options together and check each one for AI exclusions. Contact us to start.
The laws and policy forms named on this page were last checked in October 2026; we recheck them as the rules and forms change.
A few useful answers
Common questions
Does tech E&O cover AI hallucinations?
A tech E&O policy can answer for a customer’s financial loss when your AI product gives a wrong output, as long as the policy’s description of your services includes the product and no AI exclusion removes it. Ask for wording that names your AI products.
If we build on another company's model, do we still need our own insurance?
Yes. Your customers contract with you, so their claims come to you. Model providers’ terms usually cap what the provider will answer for, which leaves your own tech E&O, cyber and media coverage to respond.
How do I know if my policy has a generative AI exclusion?
Look at the endorsements listed on the declarations page for forms such as CG 40 47, CG 40 48 or CG 35 08, or for an insurer’s own AI wording. Send us your renewal and we will read it with you.
Does insurance cover copyright claims over the data used to build our model?
Media liability within a tech E&O policy can answer for some copyright claims, but policies differ widely on data-related IP and some exclude it. Patent claims usually need separate cover. Bring your data sources to the quote so the wording can be checked against them.
Does California's SB 53 apply to my AI startup?
Only if you build frontier models above its computing threshold. Most startups that build on existing models fall outside it, though the dataset disclosure rule in Civil Code section 3111 can still apply to anyone who releases or substantially modifies a generative AI system for Californians.