The agent booked the wrong thing, the model made it up, the output copied someone. Since January 2026, which policy pays is a question with an answer.
AI liability insurance covers financial loss caused by an AI system your business built, sells or runs: a hallucinated answer a customer relied on, an agent that took an action nobody authorized, output that infringed or defamed, a model that drifted. It arrives three ways, as an endorsement on tech E&O or cyber, as an add-on to a package policy, or as a standalone policy with limits to $50,000,000, and it exists because the standard forms started excluding the same thing the same year.
Tell us the situation.
One licensed human replies the same business day — not five agents, not an auto-responder. If the policy you have is already the right one, we will tell you.
We use this to quote and service your insurance, and we do not sell it or pass it to lead networks. Privacy policy.
Send us what you have. We’ll tell you if it’s right.
Your current policy paperwork, a renewal notice, or nothing at all. A licensed agent reads it and replies the same business day.
What AI liability actually covers.
Hallucination and erroneous output
A customer or client loses money relying on an answer, a recommendation, a summary or a document the AI produced that was wrong.
Without it — Tech E&O covers negligent performance of a technology service, and a 2026 renewal may carry an exclusion for anything arising out of generative AI. The claim is the same; the answer changed.
Agent actions
An autonomous agent that transacts, books, cancels, deletes, sends or approves something it should not have, for a customer or inside your own systems.
Without it — Whether the action was a security failure (cyber), a professional failure (E&O) or a product failure (products liability) is an argument between three carriers, and an argument is not a payment.
Model failure and drift
Degrading performance, a model that stops meeting the accuracy it was sold on, or an algorithmic failure across many customers at once.
Without it — Standalone AI policies were built for this peril; conventional forms were not, and a warranty of performance is excluded on most of them.
Discrimination and regulatory defense
A claim that the model’s decisions were biased, or a regulatory proceeding under a state AI or consumer law.
Without it — Colorado’s AI Act and a wave of 2026 state bills create private rights of action with statutory damages; EPLI and D&O carriers are filing AI exclusions at the same time.
IP infringement and defamation by output
Content, code or images the system generated that reproduced protected work or stated a falsehood about a real person.
Without it — These are media and IP perils, and the media form may now exclude AI. The affirmative AI endorsements bring them back under one policy.
Vendor and customer contract compliance
The limits, the AI-specific wording and the indemnity backing an enterprise customer or a platform partner requires before signing.
Without it — Enterprise procurement now asks for AI coverage by name. A tech E&O certificate that is silent on AI answers a question the customer stopped asking.
The claims we actually see.
Ordered by how often we see them, not by how dramatic they are. Each one names the coverage that answers it — and the policy people wrongly assume already does.
Most common
A customer relies on a wrong answer from your chatbot or copilot and loses money
Covered by Tech E&O with affirmative AI wording, or a standalone AI liability policy
The developer of the model won Walters v. OpenAI because it warned users the output could be wrong. Your business, which deployed it and presented the answer as a service, gets no such defense.
Common
An agent with a payment or admin credential does something it was not asked to
Covered by Cyber with an affirmative AI endorsement for a security failure; AI liability for a performance failure
Coalition and Beazley have written endorsements saying an agent’s failure is covered like a human’s. Carriers that have not said so will litigate which policy it was.
Less common, severe
A rights holder claims the output copied their work, or the training data was theirs
Covered by Media, tech E&O or AI liability, depending on the endorsement
AI-related suits rose 978% from 2021 to 2025; copyright is 11.2% of them. The general liability and media forms that covered this by silence can now exclude it by endorsement.
Rare, catastrophic
A regulator or a class action alleges the model discriminated
Covered by AI liability with regulatory defense; some EPLI and D&O forms
W.R. Berkley’s absolute AI exclusion for D&O, E&O and fiduciary lines removes any claim arising from the use of AI. Read the management liability renewal as carefully as the tech one.
What the policy will not cover.
Pick a procedure to see what the exclusion actually means and whether it can be placed somewhere else.
The standalone forms are new and each is different; these are the exclusions that appear on most of them and on the endorsements to conventional policies.
What AI liability insurance excludes
Bodily injury and property damage
Most AI liability policies answer financial loss only. Injury from an AI-controlled physical system belongs to general liability or products liability, where ISO CG 40 47 and CG 35 08 can now exclude it.
Can it be placed? Yes, but as a separate conversation with the general liability carrier, and HSB’s SMB add-on is written for exactly this gap.
Undisclosed models and uses
The application is a schedule of what you run and where. A system not on it, or a use outside what was described, is outside the policy.
Can it be placed? Yes, by disclosure. The inventory of AI in the business is the application, and updating it mid-term is an endorsement.
Performance warranties
A guarantee of accuracy, uptime or results in a contract is excluded unless the policy specifically covers model performance, which only the standalone forms do.
Can it be placed? Partly. Armilla’s standalone policy covers degradation against a measured baseline; conventional E&O will not.
Intentional acts and known defects
Deploying a model you knew produced the failure, or ignoring a flagged defect, is excluded on every form.
Can it be placed? No. Testing records and incident logs are what turn a claim from excluded to covered.
Fines and penalties where uninsurable
Regulatory penalties are covered only where the law allows, which varies by state and by statute.
Can it be placed? Partly. Defense costs are usually covered even where the fine is not.
What clients with AI liability ask us.
What is AI liability insurance?
Coverage for financial loss caused by an artificial intelligence system your business builds, sells or uses: wrong output a customer relied on, an autonomous agent’s unauthorized action, infringing or defamatory content it generated, a model that stopped performing, or a discrimination or regulatory claim over its decisions. It is sold as an endorsement on technology E&O and cyber, as an add-on to a business policy, or as a standalone policy with limits up to $50,000,000 for AI vendors.
Doesn’t my tech E&O or cyber policy already cover AI?
Until 2026 it often did by silence, because the forms never mentioned AI. In January 2026 ISO published generative-AI exclusions for general liability, W.R. Berkley filed an absolute AI exclusion across D&O, E&O and fiduciary lines, and Hamilton filed one naming ChatGPT and Midjourney. Other carriers went the other way: Beazley said it will not exclude AI, Hiscox and Coalition added affirmative AI wording. The honest answer is on your declarations page, and the safe one is to ask for the affirmative wording.
Who actually writes AI liability insurance?
Standalone: Armilla with Chaucer at Lloyd’s (to $25,000,000, requiring a model assessment), AIUC on Beazley paper (to $50,000,000 of product liability for AI vendors and their customers, underwritten on a technical certification), Munich Re’s aiSure through Mosaic (to $15,000,000), and Corgi and Vouch for startups. Endorsed onto existing lines: Coalition on cyber and tech E&O, Beazley’s AI cyber endorsement, Hiscox in its technology book, and HSB’s AI liability add-on sold through partner carriers to small businesses.
What does agentic AI insurance cover that ordinary AI coverage does not?
The action, not just the answer. A chatbot that says something wrong causes a loss when someone relies on it; an agent with a credential causes the loss itself, by paying, booking, deleting or approving. Coverage for agents has to say that an autonomous action is treated like an employee’s, which is what Coalition’s endorsement and Beazley’s AI cyber endorsement do, and it has to say whether the loss is yours or your customer’s. AIUC’s product is written for vendors whose agents act inside customers’ systems.
What does AI liability insurance cost?
An AI endorsement on an existing tech E&O or cyber policy is usually a modest addition to a premium that for a small software company runs from about $2,000 to $10,000 a year at $1,000,000 of limit. Standalone policies are priced on the model’s use, its testing evidence and the limit, and carriers do not publish rates; expect a technical review before a number. An AI vendor whose enterprise customers require $5,000,000 or more should budget for a real underwriting process rather than an online quote.
What will the underwriter ask?
An inventory of every AI system in the business and what each one does. Which models, whether you trained them and on what, and the vendor terms behind third-party ones. Where a human reviews output before it reaches a customer. What the agent is allowed to do without approval, and what credentials it holds. Testing, monitoring and incident records. Your customer contracts, especially indemnities and performance promises. The businesses that get quoted are the ones that can hand this over in a day.
We are a small business that uses AI tools. Do we need a separate AI policy?
Usually not a separate one. A business that uses AI to write, answer customers, summarize or analyze needs its existing policies to cover what those tools do: media liability if it publishes, tech E&O if it sells a service, cyber for the data, and general liability without a generative-AI exclusion. HSB’s AI liability add-on was built for that buyer and attaches to a business policy through the carrier. A standalone policy is for companies whose product is the AI.
Commercial insurance is 96% of what we do — it is not a department here, it is the whole business. Not ready to talk? The guides answer the questions this page raises in more depth. Already insured with us and need a certificate or a policy change? Ask the service team rather than starting a quote — it is faster and it goes to the people whose job it is. We also write home and auto, which is usually cheaper alongside the business policy than apart from it.
Handed a lease, a vendor packet or a client contract with insurance requirements? Paste it into the COI Requirements Decoder to see what each one means for your business, or read what the requirements mean.