AI Company Insurance:
Tech E&O, Cyber, and the Coverage Gaps Specific to AI Products
Artificial intelligence (AI) company insurance is a technology errors and omissions (tech E&O) and cyber program adapted for the specific ways AI products fail — model errors, biased outputs, intellectual property infringement in training data, and regulatory investigations. The exposure is real and growing, but standard policies increasingly exclude it: in 2026, insurers began formally writing generative-AI exclusions into general liability, directors and officers (D&O), and E&O forms. For AI companies, the coverage that matters is affirmative AI coverage that does not assume the gap away.
Photo by Israel Andrade
on Unsplash
- AI products fail in ways generic software does not — model errors and hallucinations, biased or discriminatory outputs, IP infringement tied to training data, and regulatory investigations — and each maps to a different coverage line.
- The insurance market is now codifying the AI gap as explicit exclusions. In 2026, the Insurance Services Office (ISO) introduced optional generative-AI exclusion endorsements (forms CG 40 47 and CG 40 48) for general liability, with parallel language spreading into D&O and E&O forms.
- The counter-move is affirmative AI coverage — Coalition, Embroker, Relm, and Munich Re have all launched AI-specific endorsements or policies since 2024 that say, in writing, that AI-related events are covered.
- AI incidents are rising fast: the AI Incident Database recorded 362 incidents in 2025, up from 233 in 2024, per Stanford's AI Index — the kind of loss data underwriters use to tighten or price AI risk.
- For an AI company, the job is not buying a single "AI policy." It is making sure your tech E&O, cyber, and management-liability forms affirmatively cover AI — and that the coverage matches the AI-specific clauses now appearing in customer contracts.
What "AI company insurance" actually means
AI company insurance is not a single product. It is a technology errors and omissions (tech E&O) and cyber program — usually with directors and officers (D&O) and general liability alongside — adapted to cover the distinct ways AI products cause loss. What makes AI its own risk class is that the product can be wrong in ways its makers cannot fully predict or audit, and the resulting harm can far exceed the contract value.
If you build, train, fine-tune, or embed AI models in a product you sell, your exposure looks different from a conventional software company's. A scheduling tool that crashes triggers a fairly well-understood tech E&O claim. An AI model that produces a confidently wrong answer, a biased decision, or output that infringes someone's copyright triggers a claim that standard forms were never written for. For the full coverage picture behind these risks, start with our complete guide to tech and SaaS insurance; this page focuses specifically on what changes when the product is AI. Most AI accounts trace back to the operations on our SaaS and software insurance hub.
The exposure is not theoretical. Stanford's AI Index, drawing on the AI Incident Database, recorded 362 documented AI incidents in 2025 — up from 233 in 2024, itself a 56% jump over the prior year. Those incidents span privacy violations, discriminatory outputs, misinformation, and consequential algorithmic errors — exactly the categories that turn into claims, regulatory inquiries, and lawsuits.
Why standard tech E&O and cyber policies miss AI exposure
For years, AI claims fell into a gray area — arguably covered, arguably not, under policies written before generative AI existed. In 2026, insurers started removing the ambiguity in the direction that protects them: by adding explicit AI exclusions. A tech E&O or cyber policy that does not affirmatively address AI may now actively exclude it, which means an AI company can be paying for coverage that carves out its single biggest exposure.
The clearest signal came from the Insurance Services Office (ISO), the body whose standardized forms most U.S. carriers build on. ISO introduced optional generative-AI exclusion endorsements — forms CG 40 47 and CG 40 48 — for commercial general liability (CGL) policies. CG 40 47 is broad, barring coverage under both bodily-injury/property-damage and personal-and-advertising-injury for harms tied to generative-AI output; CG 40 48 is narrower, excluding only the personal-and-advertising-injury coverage where AI-generated defamation or IP infringement would otherwise land. Industry analyses report parallel exclusion language now spreading into D&O and management-liability and E&O forms, with carriers including AIG, W.R. Berkley, and Great American seeking regulatory clearance for AI-specific management-liability exclusions.
This is why "do I have AI coverage?" is rarely answered by the policy's headline. The answer lives in the endorsements. A company can hold a perfectly good technology errors and omissions (tech E&O) policy and a strong cyber liability policy and still have AI carved out by a one-page exclusion stapled to the back. Some analysts expect AI exclusions to become a default feature of many policy forms within a couple of years unless buyers push back — which makes reading the forms, not the brochure, the only reliable check.
The five AI-specific exposures underwriters now scrutinize
AI underwriters break the risk into five exposures that conventional software does not carry in the same way: model errors and hallucinations, biased or discriminatory outputs, intellectual property (IP) infringement, regulatory investigations, and AI-enabled cyber fraud. Each maps to a different coverage trigger, which is why a single "AI policy" rarely covers the whole picture — the exposures sit across tech E&O, cyber, and management liability.
The framework below mirrors how AI-focused brokers and carriers categorize the risk. The point is not that every AI company faces all five equally — a developer-tools startup and an AI hiring platform have very different profiles — but that each exposure needs a named home in the program, not an assumption that "the E&O covers it."
| AI exposure | What triggers a claim | Coverage line that should respond |
|---|---|---|
| Model error / hallucination | The model produces a confidently wrong output a customer relies on, causing financial loss beyond the contract value | Tech E&O (with affirmative AI language) |
| Bias & discrimination | Algorithmic decisions in hiring, lending, or healthcare produce discriminatory outcomes, drawing class actions or penalties | Tech E&O / specialized bias coverage; sometimes employment practices liability |
| IP infringement | Training data or generated output is alleged to infringe copyright, trademark, or patent | Tech E&O / media or IP coverage (and the gap CGL exclusions now target) |
| Regulatory investigation | A regulator opens an inquiry under data-protection or AI-transparency law (for example, the EU's General Data Protection Regulation or California's privacy statutes) | Cyber / regulatory-defense coverage; D&O for management exposure |
| AI-enabled cyber fraud | Deepfake-enabled funds-transfer fraud (FTF) or an AI-driven security failure | Cyber liability (with affirmative AI / deepfake language) |
The IP and bias exposures are the ones AI founders most often underestimate. Generative-AI products have drawn a wave of copyright litigation over training data, and AI used in consequential decisions — lending, hiring, healthcare — sits squarely in anti-discrimination law. Both are areas where the new CGL and E&O exclusions bite hardest, because defamation and IP infringement from AI output are exactly what forms like ISO's CG 40 48 are written to exclude.
How AI coverage is actually being written in 2026
As insurers added AI exclusions, a parallel market for affirmative AI coverage emerged — endorsements and policies that state, in writing, that AI-related events are covered rather than carved out. For most AI companies the cleanest path is not a standalone "AI policy" but a tech E&O and cyber program from a carrier that has built affirmative AI language into its forms.
The affirmative side of the market has moved quickly since 2024:
- Coalition added an Affirmative Artificial Intelligence Endorsement to its cyber policies in March 2024, expanding the breach trigger to include an AI-caused security failure and extending funds-transfer-fraud coverage to deepfake-enabled fraudulent instructions. Its Active Cyber Policy (issued on all new and renewal U.S. surplus-lines business from April 15, 2025) builds that affirmative AI coverage in, and a December 2025 Deepfake Response Endorsement added dedicated deepfake-incident response.
- Embroker added an Artificial Intelligence Coverage Endorsement effective August 5, 2025, included on every technology E&O and cyber quote rather than sold as a separate buy.
- Relm Insurance launched three AI-specific policies in January 2025 — NOVAAI (cyber and tech E&O for AI companies), PONTAAI (an excess difference-in-conditions wrap for organizations whose existing policies exclude AI), and RESCAAI (a first-party response policy for businesses using third-party AI).
- Munich Re has written AI performance-guarantee coverage through its aiSure program since 2018, indemnifying clients for financial losses caused by underperforming AI models; in 2025 it partnered with specialty insurer Mosaic to extend AI coverage to AI vendors.
Pricing depends on revenue, the criticality of the AI system, data sensitivity, and governance maturity — companies that can show model documentation, bias testing, and risk controls underwrite better. Specialized E&O or cyber coverage for a startup commonly runs in the low thousands per year; one insurtech broker reports a median cyber premium of $2,968 for startups, with AI, fintech, and healthtech accounts paying more. A full multi-line technology program more typically lands anywhere from $5,000 to $250,000+ per year, or roughly 0.5%–3% of revenue, depending on lines and limits. For how those numbers break down across the whole program, see our tech and SaaS insurance cost guide, and for which carriers write what, our tech and SaaS carrier market guide.
What to check before you bind — and before you sign a customer contract
For an AI company, the two documents that decide whether you are actually covered are your policy's AI endorsement and your customer's contract. The trap is a mismatch: a customer master service agreement (MSA) that requires you to carry insurance for AI errors, paired with a policy that quietly excludes them. Catching that gap is a form-reading exercise, not a price comparison.
Three checks matter most before binding:
- Read the endorsements, not the declarations. Confirm whether the tech E&O and cyber forms carry an affirmative AI grant or an AI exclusion. The headline policy can look identical between a carrier that covers AI and one that excludes it — the difference is a one-page endorsement.
- Match limits to contracts. Enterprise customers routinely require $1M–$5M in tech E&O and cyber limits, and AI vendors are increasingly seeing contract clauses that specifically demand coverage for AI or algorithmic errors. Your contracts often set the floor before any underwriting question is asked. Our tech and SaaS requirements guide walks through the contractual and investor standards these forms must satisfy.
- Plan management liability around funding. If you are venture-backed, investors typically expect general liability and a business owner's policy (BOP) as a baseline, plus directors and officers (D&O) of $3M–$5M placed within 60–90 days of a round. AI-specific exclusions are now appearing in D&O forms too, so the same "affirmative or excluded?" question applies there.
This is where an independent broker earns the relationship on an AI account: the same submission can come back from one carrier with affirmative AI language and from another with an AI exclusion at a nearly identical price, and only a line-by-line form comparison reveals which "yes" actually protects you. If a claim does hit, our tech and SaaS claims guide covers the response sequence; the cyber dynamics overlap with other data-heavy verticals, as our hotel cyber insurance guide shows.
The AI exclusion that would have put a vendor in breach of its own customer contract
An early-stage AI company we work with sells a document-analysis tool into large enterprises. They came to us not for a claim but for a contract: a major customer's master service agreement required them to carry technology errors and omissions and cyber coverage that specifically extended to "errors or omissions arising from artificial intelligence." They already had a tech E&O and cyber quote in hand from a well-known program and were a day from binding it. On paper it looked fine — the right limits, a competitive price.
The problem was on the last page. The quote carried a recently added generative-AI exclusion endorsement. Had they bound it and signed the MSA, they would have been in immediate breach of the contract's insurance clause — and worse, the exact exposure the customer was worried about, a faulty AI output, would have fallen straight into the gap. We did not find that by comparing premiums; we found it by reading the endorsement against the contract clause side by side. We re-marketed the account to a carrier whose tech E&O and cyber forms carried affirmative AI language, matched the limits the MSA required, and got them bound in time to sign. The premium difference was modest; the coverage difference was the whole deal. The lesson we take to every AI placement: on an AI account, the endorsement page decides whether you are covered, and the brochure will never tell you.
Details anonymized and generalized to protect client confidentiality.
Frequently asked questions about AI company insurance
Most AI companies need a technology errors and omissions (tech E&O) and cyber program at the core, plus general liability, a business owner's policy (BOP), and — if venture-backed — directors and officers (D&O). What makes AI different is the exposures these forms have to reach: model errors, biased outputs, intellectual property (IP) infringement, regulatory investigations, and AI-enabled fraud.
The key is making sure each line affirmatively covers AI rather than excluding it. See our tech and SaaS insurance FAQ for answers across the rest of the program.
Not necessarily — and increasingly, not. In 2026, insurers began adding optional generative-AI exclusion endorsements to general liability, E&O, and D&O forms. A policy can look complete on its declarations page and still carve out AI through a one-page endorsement, so the only reliable answer comes from reading the actual endorsements, not the brochure.
Affirmative AI coverage is policy language that states, in writing, that AI-related events are covered rather than left ambiguous or excluded. Carriers including Coalition, Embroker, Relm, and Munich Re have launched affirmative AI endorsements or AI-specific policies since 2024. It is the opposite of "silent" AI coverage, where neither the grant nor an exclusion clearly addresses AI and a dispute is likely.
Specialized E&O or cyber coverage for an AI startup commonly runs in the low thousands of dollars per year — one insurtech broker reports a median startup cyber premium near $2,968 — with AI, fintech, and healthtech accounts paying more for the same limits. A full multi-line technology program more typically lands between $5,000 and $250,000+ per year, or roughly 0.5%–3% of revenue, depending on lines, limits, and controls.
Underwriters focus on five: model errors and hallucinations (a wrong output a customer relies on), bias and discrimination (algorithmic decisions in hiring, lending, or healthcare), IP infringement (training data or generated output), regulatory investigations (under data-protection and AI laws), and AI-enabled cyber fraud (deepfake-driven funds-transfer fraud). They sit across tech E&O, cyber, and management liability rather than in one place.
For most AI companies, an endorsement is enough — the goal is tech E&O and cyber forms that affirmatively cover AI, not a standalone product. Standalone AI policies (such as performance-guarantee or excess wrap coverage) exist and make sense in specific cases, like indemnifying a customer for an AI model's underperformance or wrapping a policy that excludes AI. Which structure fits depends on your product, contracts, and existing forms.
You need tech E&O and cyber forms whose AI language affirmatively matches the contract — and limits that meet the contract's floor, often $1M–$5M for enterprise master service agreements. The risk is binding a policy with a generative-AI exclusion while signing a contract that requires AI coverage, which puts you in immediate breach. Have the endorsements reviewed against the contract clause before you sign.
Not sure whether your policy covers AI or excludes it?
Ask about affirmative AI coverage, generative-AI exclusions, and the tech E&O and cyber forms AI companies actually need.
Have us pressure-test your AI coverage against your contracts
Tell us what your AI product does and what your customer and investor contracts require, and we'll check your tech E&O, cyber, and management-liability forms for AI exclusions — and shop carriers with affirmative AI language.