AI Liability Insurance Buyer's Guide

AI Endorsement vs AI Insurance: What's the Difference?

By Joel R. Singh · Last updated August 6, 2026

Walk into a broker conversation about AI coverage and you may hear two different things described with the same word: "endorsement." One kind adds coverage. The other strips it away. Understanding which is which, and how either compares to a dedicated standalone AI insurance policy, is the key differentiator between a program that provides coverage when something goes awry versus an exclusion that leaves you shouldering the entire loss.

This page explains what each structure is, how they differ in coverage scope, limits, and claim triggers, and a practical sequence for figuring out which one your business needs.

This is general information, not insurance advice, and it is not a recommendation of any specific policy. Talk to a licensed broker who will guide you into customizing what you need for your program.

What an AI endorsement actually is

An endorsement is a modification to a policy that already exists. It rides on a host policy, sharing that policy's limits, its premium payment cycle, and its claims-made or occurrence structure. Endorsements cannot exist on their own. Pull the host policy and the endorsement falls away with it.

The term "AI endorsement" is used for two opposite things in the market, and the distinction is not subtle.

Type 1: AI exclusion endorsements. These remove coverage. Verisk (ISO) published a suite of them effective January 1, 2026. CG 40 47 attaches to commercial general liability (CGL) policies and removes coverage for bodily injury, property damage, and advertising injury whenever those harms trace back to generative AI. A companion form, CG 35 08, applies the same exclusion to products and completed operations coverage. The ISO definition of generative AI in these forms is broad: it reaches any machine-based learning system trained to create content, including text, images, audio, video, and code. That captures every mainstream tool from ChatGPT to AI image generators to code-completion assistants. These forms are optional, not automatic, but they give every carrier standard regulator-reviewed language to attach at renewal. (See our full breakdown of CG 40 47 and what it removes.)

Type 2: Affirmative AI endorsements. These add coverage. Where a host policy is silent on AI, or where a prior exclusion has been removed by negotiation, an affirmative AI endorsement states explicitly that specified AI-related claims are covered. The coverage does not exist somewhere else in the policy by implication: the endorsement creates it. Limits are shared with the host policy, which means a large AI claim can erode the same aggregate that protects your general business operations. Scope is bounded by whatever the endorsement language says and whatever the host policy does not exclude. Read both documents, not just the endorsement.

Affirmative AI endorsements in the market today

Several carriers have introduced affirmative AI endorsements. Each works differently, and the scope differences are material.

Coalition Affirmative AI Endorsement (cyber endorsement)

Attached to Coalition's cyber policies in the US surplus and Canada markets. The endorsement expands the definition of a security failure to include an AI security event, where AI technology caused a failure of computer systems' security. It also expands funds transfer fraud triggers to include deepfake and AI-generated fraudulent instructions. The key limitation: this endorsement covers AI as an attack vector, the scenario where someone uses AI against you, not liability for harm your own AI output causes to a third party. A chatbot that gives a customer dangerous medical advice is not what this endorsement was designed for.

Embroker AI Coverage Endorsement (tech E&O / cyber endorsement)

Automatic on eligible tech E&O and cyber quotes in Embroker's Startup Program as of August 2025. Built to address risks from machine learning models, autonomous decision systems, and algorithmic service delivery, where standard tech E&O language was not drafted to respond. Target segment is technology startups; not available as a general-market standalone.

Vouch AI Coverage

Vouch packages AI coverage within its technology company product suite, covering AI errors and omissions, algorithmic bias and discrimination claims, regulatory investigation defense costs, and IP infringement claims arising from AI systems. Available via Vouch's platform, primarily for VC-backed AI startups. Note on carrier structure: Corix is Vouch's own MGA and carrier operation, and in 2025 Hiscox agreed to acquire Corix from Vouch. Post-close, Vouch continues as an independent broker with a multi-year distribution arrangement with Hiscox. Confirm the current paper when quoting, as the structure is in transition.

What standalone AI insurance is

A standalone AI insurance policy is a dedicated affirmative policy: its own limits, its own covered perils, its own trigger language, underwritten specifically for AI liability. It does not depend on a host policy to exist. Claims against the AI policy do not erode your cyber limit or your tech E&O aggregate.

The standalone market is real but early. As of mid-2026, only a small number of true standalone AI liability products exist worldwide. The products that exist differ substantially in trigger structure, target segment, and distribution path.

Armilla Affirmative AI Liability / Vanguard AI (standalone AI liability)

Launched April 2025 as among the first affirmative AI liability policies written at Lloyd's. Underwritten by Chaucer and other Lloyd's syndicates, with Armilla as coverholder. Coverage triggers on AI underperformance: hallucinations, model drift, mechanical failures, and critical errors leading to damages. Armilla requires an AI system assessment before it will write a policy. Limits of $25 million or more available per organization. In early 2026, Armilla and Chaucer signaled Vanguard AI, a coordinated structure pairing AI liability limits with cyber limits under predefined allocation rules for mixed scenarios. Available to US-based insureds through surplus lines brokers.

Munich Re aiSure (via Mosaic Insurance) (AI performance guarantee)

A parametric-style structure: no negligence allegation required. A breach of a predefined performance threshold triggers payment, and claims settle on measurable performance data without lengthy investigation. Mosaic partnered with Munich Re to distribute aiSure starting February 2026, with up to $15 million in initial capacity per insured (available in multiple currencies, including euros and Canadian dollars). Coverage addresses bias failures, privacy failures, IP infringement, and performance shortfalls. Suited to AI vendors who make contractual performance promises to their customers and need insurance to backstop those commitments.

Testudo GenAI Liability (standalone AI liability)

Launched January 2026 as a managing general agent (MGA) backed by Lloyd's (Apollo, Atrium, QBE syndicates). Capacity expanded to $9.25 million* per insured by March 2026; in August 2026 we received information from Testudo that the limit was increased to $10 million per insured. Explicitly designed to fill the gap left by CG 40 47 exclusions. Covers third-party claims from AI-generated outputs including hallucinations, model drift, IP infringement, unauthorized data disclosure, bodily injury, property damage, and regulatory proceedings. US focus, mid-market segment.

HSB AI Liability Insurance for Small Businesses (standalone AI liability, SMB)

Announced March 2026 by HSB, a Munich Re subsidiary. The only verified product in this market with an explicit small-business target and broad US state availability. Not sold direct: delivered as an addition to business policies of carriers that partner with HSB, pending state regulatory approval in each state. Covers lawsuits arising from AI use, including bodily injury, property damage, and advertising injury from AI-generated marketing, blogs, and social media content.

Relm PONTAAI (excess DIC wrap)

Built specifically for businesses whose existing program already carries an AI exclusion. PONTAAI is an excess difference-in-conditions wrap: it sits over existing CGL, tech E&O, and cyber policies and responds where those policies do not, filling the hole left by AI exclusion endorsements. A third-party claim that falls outside your primary program because of CG 40 47 or a carrier-specific exclusion may drop down to PONTAAI. Available through brokers to Relm's Bermuda specialty market; verify admitted versus surplus lines status for your state.

How coverage, limits, and triggers differ

Affirmative AI endorsement Standalone AI policy
Limits Shared with host policy. An AI claim erodes the same aggregate as all other covered claims. Dedicated AI aggregate. AI claims do not erode cyber, E&O, or CGL limits.
Claim trigger Inherits the host policy trigger: claims-made for most cyber and E&O; occurrence for CGL. Policy-specific. Some (Armilla, Testudo) use a fault/negligence trigger; Munich Re aiSure uses a parametric performance-threshold trigger.
Covered perils Bounded by the host policy category. May not reach bodily injury or advertising injury if the host policy does not. Purpose-built perils: hallucinations, model drift, IP infringement, bias failures, regulatory proceedings, advertising injury, and bodily injury can be written as first-class perils.
Underwriting Typically streamlined; host underwriting already assessed the business. May add an AI usage questionnaire. More involved. Armilla requires an AI system assessment before binding. AIUC requires certification against its AIUC-1 standard across 5,000+ adversarial tests.
Distribution Via the broker who places the host policy. Some platforms (Embroker) include it automatically. Surplus lines broker for most products. Not available off the shelf.

How to tell which you have and which you need

Pull the endorsement schedule from every policy in your program.

Look for four things: (a) CG 40 47 or CG 35 08; (b) any carrier-specific AI exclusion with "artificial intelligence" in the title; (c) any endorsement described as "affirmative" or naming specific AI perils as covered; (d) any buy-back or write-back endorsement that explicitly restores AI coverage the base policy removed. Do not rely on a verbal summary. Get the form numbers in writing from your broker.

Map your AI exposure against what the endorsement actually covers.

A cyber endorsement that covers AI security events does not cover third-party harm from your AI's output. A tech E&O endorsement that covers AI professional errors may not respond to bodily injury. Write down your actual exposures: Does your business publish AI-generated marketing content? Run a customer-facing chatbot? Use AI in medical, legal, or financial advice workflows? Then check whether your endorsement language specifically addresses those scenarios. Gaps between your exposure and the endorsement's scope are uninsured.

Ask your broker whether a buy-back is available before going to a standalone market.

Some carriers will attach an affirmative AI endorsement to restore coverage they just excluded, at an additional premium. Staying within the existing policy means no new underwriting relationship, no additional surplus lines broker, and no separate limit to manage. If no buy-back is available, or if the buy-back scope does not address your exposures, that is the signal to price a standalone policy.

Build the AI governance documentation before you shop.

Whether you end up with an endorsement or a standalone policy, underwriters are moving toward governance-based pricing: model inventories, documented human oversight, incident response procedures, and bias mitigation records. Organizations that can produce this documentation get workable terms; those that cannot face exclusions, sublimits, or declined quotes. Build the documentation set once, use it across every renewal. (Our coverage-readiness checklist documents what underwriters actually ask for.)

When each approach makes sense

An affirmative AI endorsement is typically the right starting point when

  • AI is a feature of the business, not the primary product being sold
  • The existing policy structure matches the primary exposure (cyber host for AI security events; tech E&O host for AI professional errors)
  • AI claims are unlikely to be large enough to exhaust shared limits
  • The host carrier offers a scope-appropriate buy-back at renewal

Standalone AI insurance makes more sense when

  • AI output is central to what the business sells or delivers
  • Contracts require dedicated AI liability limits, separate from cyber or E&O
  • The existing program now carries a broad exclusion and no buy-back is available
  • The exposure includes bodily injury, advertising injury, or regulatory proceedings that the host policy category does not cover
  • AI claims are large enough in probability or severity to warrant a dedicated aggregate

The two approaches are not mutually exclusive. Some programs carry both: an endorsement on an existing policy for the coverage category that policy handles well, and a standalone policy for the exposures that fall outside any existing structure.

See also: the carrier comparison table (every verified product across both categories), "Does E&O Cover AI?", and the CG 40 47 explainer.

Sources

  1. Verisk to Roll Out New GL Exclusions for Generative AI Exposures — Big I Virtual University
  2. The End of Silent AI? Emerging AI Exclusions, Coverage Fragmentation, and Practical Implications — Fenwick
  3. Coalition Adds New Affirmative AI Endorsement to Cyber Policies — Coalition
  4. Introducing AI Coverage for Tech Companies — Embroker
  5. Armilla Launches Affirmative AI Liability Insurance with Lloyd's Underwriter, Chaucer — Armilla / PR Newswire
  6. Mosaic Partners with Munich Re's aiSure to Provide Pioneering Coverage for AI Vendors — Mosaic Insurance
  7. Introducing AI Liability Insurance for Small Businesses — HSB / Munich Re
  8. Relm's PONTAAI Solution — Relm Insurance
  9. Testudo GenAI Liability — Testudo / The Insurer
  10. Hiscox to Acquire MGA and Carrier Operations from Vouch — Hiscox Group / Coverager
  11. AIUC-1 Standard Explained — Mindgard / AIUC
  12. The First AI Liability Insurance Product Has $25 Million in Coverage — Medium / Purdy House

This page is general information about the insurance market, not insurance, legal, or financial advice. Coverage terms vary by carrier, state, and policy. Consult a licensed insurance broker about your specific situation. Last updated: August 6, 2026.

HOME