The AI Coverage Buyer's Guide presents
The AI Coverage
Buyer's Brief
A source-verified brief on where AI liability coverage stands in the US: what is being excluded, what you can still buy, and how to walk into a renewal ready.
Author
Joel R. Singh
Edition
Version 1.0 · 2026 Edition · Updated 2026-07-12
Placement policy
Not for sale
Section 01
Executive summary
Two things happened to AI liability coverage in 2026 at the same time. Standard policies started carving AI out, and a small standalone market grew up to sell it back. Here is the landscape in five points.
The exclusion is now standard. On January 1, 2026, Verisk/ISO forms CG 40 47 and CG 35 08 took effect. Where a carrier attaches them, your general liability policy no longer responds to injury or damage arising from generative AI. verified 2026-07-12
It goes wider than the ISO forms. At least six major carriers have filed their own AI exclusions, and WR Berkley has an absolute AI exclusion sweeping D&O, E&O, and fiduciary lines. Roughly 80% or more of these filings have been approved by state regulators.
A real standalone market exists, but it is small. Only a handful of true standalone AI liability products are on the US market as of mid-2026, most surplus lines and Lloyd's or Munich Re backed. Everything else is an endorsement on cyber or tech E&O.
Underwriting is now a documentation exam. Carriers increasingly price and condition coverage on your AI governance: model inventories, human oversight records, bias testing, and incident response. What you can produce determines your terms.
The gap lands at renewal, quietly. The exclusion arrives as an endorsement on an otherwise routine renewal. If nobody reads the form numbers, the first time you learn AI is excluded is when a claim is denied.
What to do this quarter
Pull your current GL, E&O, D&O, and cyber policies and search each for the letters "AI" and for the form numbers CG 40 47 and CG 35 08. Before your next renewal, assemble the nine documents in Section 05 and take the Section 06 script to your broker. Getting ready is cheaper than a denied claim, and it is the single thing most under your control.
Section 02
The exclusion wave
The center of the 2026 shift is two standardized endorsements. They are short, they are boring to read, and they change what your general liability policy does.
Commercial General Liability
Excludes bodily injury, property damage, and personal and advertising injury arising from generative AI. If your carrier attaches this at renewal, harm your business causes through generative AI (for example AI-written marketing that defames someone) is no longer covered by your CGL.
effective 2026-01-01Products and completed operations
The products and completed-operations counterpart to CG 40 47. If your product or completed work involves generative AI and causes injury or damage after delivery, this endorsement removes that coverage. Anyone shipping an AI-enabled product should check renewals for this form number.
effective 2026-01-01The wording can sweep in nearly any modern business activity: "any actual or alleged use, deployment, or development of Artificial Intelligence."
The broadest one: WR Berkley's absolute exclusion
Beyond the ISO forms, WR Berkley has introduced an absolute AI exclusion on D&O, E&O, and fiduciary lines. The wording above is broad enough that management decisions about AI, professional services delivered with AI assistance, and AI-related shareholder claims may all fall outside coverage. If this lands on your D&O or E&O renewal, push back through your broker or seek alternative markets.
The six exclusions we track
| Issuer | Form | Lines affected | Status |
|---|---|---|---|
| Verisk / ISO | CG 40 47 | Commercial General Liability (generative AI) | verified |
| Verisk / ISO | CG 35 08 | Products and completed operations | verified |
| WR Berkley | Absolute AI exclusion | D&O, E&O, fiduciary | verified |
| Chubb | Carrier filings | Across commercial lines | needs primary verification |
| Industry aggregate (6+ carriers) | Various endorsements | Across commercial lines | needs primary verification |
| Lawyers professional liability carriers | Manuscript exclusions | Lawyers professional liability | verified |
Full source links, plain-English reads, and a dated changelog for each filing live in the AI Exclusion Tracker.
Section 03
Does your E&O or CGL actually cover AI?
Before 2026, standard errors and omissions and general liability policies quietly absorbed a lot of AI exposure, because nobody had written AI out yet. A hallucinated recommendation, a biased screening tool, an AI-drafted document that caused a client loss, these usually fell inside the broad wording of an E&O or professional liability policy.
The 2026 exclusions remove exactly that silent coverage. Once CG 40 47, CG 35 08, or a carrier manuscript exclusion attaches, the activity you most need covered is the activity the endorsement names.
The first time many businesses learn AI is excluded is when a claim is denied. Read the form numbers before the renewal, not after the loss.
Three paths to close the gap
Once you know what is excluded and what your business actually uses, there are broadly three paths.
Negotiate the exclusion at renewal
Sometimes possible while the market is still competitive. Ask the carrier to narrow or remove the exclusion, or to add a buy-back endorsement. Governance documentation strengthens your hand.
Add an AI endorsement
Attach affirmative AI coverage to an existing cyber or tech E&O policy. The most common route today; check whether it covers your AI's outputs or only AI as an attack vector.
Buy a standalone AI policy
A dedicated AI liability product or a DIC wrap sitting over your excluded program. Mostly surplus lines; expect an AI system assessment before a carrier will write it.
The full decision framework, including who is most exposed, is in Does E&O Cover AI?
Section 04
The carrier landscape
We track 12 AI liability products across four shapes: true standalone policies, performance guarantees, DIC wraps that sit over excluded programs, and endorsements on cyber or tech E&O. A summary of the market is below; the full live table with a source link and last-verified date on every cell is at /comparison.html.
| Product | Type | Target | How to buy |
|---|---|---|---|
| Armilla Vanguard AI | Standalone AI liability | Enterprise AI vendors and deployers | Surplus lines broker; AI assessment required |
| Munich Re aiSure | AI performance guarantee | AI developers and vendors | Broker to Mosaic or Munich Re |
| Testudo GenAI Liability | Standalone AI liability | Mid-market US vendors and deployers | Broker placement via Testudo (MGA) |
| HSB AI Liability for SMB | Standalone AI liability (SMB) | Small and mid-sized businesses | Added to partner-carrier policies |
| AIUC AI Agent Insurance | Vendor-bundled with certification | Enterprises deploying AI agents | Via AIUC or bundling vendors; AIUC-1 cert |
| Vouch AI Insurance | Cyber / E&O endorsement tier | VC-backed AI startups | Direct via Vouch |
| Relm NOVAAI | Cyber + tech E&O for AI companies | AI platform and product companies | Broker to Relm (Bermuda) |
| Relm PONTAAI | Excess DIC wrap for AI exclusions | Deployers whose programs exclude AI | Broker to Relm (Bermuda) |
| Relm RESCAAI | First-party AI loss coverage | Companies with first-party AI losses | Broker to Relm (Bermuda) |
| Coalition Affirmative AI | Cyber endorsement (attack vector) | Cyber policyholders (US surplus, Canada) | Via Coalition cyber policy |
| Embroker AI Endorsement | Tech E&O / cyber endorsement | Startups (Embroker Startup Program) | Direct; automatic on eligible quotes |
| CFC Tech E&O | Tech E&O (AI positioning unverified) | Not disclosed / verify with carrier | Broker to CFC |
A note on reading this market: capacity figures changed three times in six weeks during compilation. Treat any specific limit as a point-in-time reading and confirm current terms with the carrier or a broker. The full comparison table carries the dated source on every cell.
Section 05
The coverage-readiness checklist
AI underwriting has become a documentation exam. These are the nine documents underwriters ask for, in the order it makes sense to build them. Organizations that can produce this set get coverage on workable terms; those that cannot face exclusions, sublimits, or premium increases.
- 01
AI systems inventory
You cannot price, govern, or insure what you have not enumerated. Every other document refers back to this one, so it is where to start.
- 02
AI acceptable use policy
The fastest signal of whether AI usage is governed at all. No policy means, by definition, ungoverned usage, and the underwriter assumes the worst case.
- 03
Governance framework with board-level oversight
Accountability that stops at the IT department signals AI risk is treated as a technical issue rather than a business one.
- 04
Risk register with AI failure modes
A business that has written down "our chatbot could give a customer harmful advice" and mitigated it is a different risk from one that never asked.
- 05
Bias testing and validation records
Discrimination and bias claims are among the perils AI-specific products are built to cover, which tells you where carriers expect losses.
- 06
Vendor due diligence files and AI contract provisions
Most organizations' AI risk is vendor risk. Underwriters want to know whether a vendor failure lands on you or is contractually pushed back.
- 07
Incident response plan covering AI incidents
Claims cost depends on response speed. A business that can detect, stop, and document an AI failure is a materially better risk.
- 08
Employee training records
Policies without training are shelfware. Training records are the evidence your acceptable use policy exists in practice, not just on paper.
- 09
Human oversight documentation for consequential decisions
The sharpest question of all: for the decisions that can really hurt someone, was a human in the loop, and can you prove it?
Each document, with the source behind why underwriters ask for it and a self-scoring guide, is in the Coverage-Readiness Checklist.
Section 06
The broker conversation script
Bring these questions to your next renewal. They surface an AI exclusion before it becomes a denied claim, and they tell your broker you have done the homework.
- Q1
"Does my renewal attach CG 40 47, CG 35 08, or any carrier-specific AI exclusion? Please point me to the exact endorsement form numbers on the policy."
- Q2
"Across all my lines, GL, E&O, D&O, and cyber, which ones now exclude AI, and what specifically is excluded on each?"
- Q3
"If AI is excluded, can we negotiate a narrower exclusion or a buy-back endorsement? What documentation would improve those terms?"
- Q4
"For any affirmative AI coverage you offer, does it respond to liability for my own AI's outputs, or only to AI as an attack vector against me?"
- Q5
"What standalone AI liability products or DIC wraps can you access for a business my size, and what would each carrier need to see to write it?"
- Q6
"Which of the nine governance documents would most improve my pricing or availability, and are any of them a hard requirement to bind?"
Section 07
Methodology and how this guide is funded
How the data is built
- Per-cell sourcing. Every factual claim in our datasets links to its source and carries a last-verified date. Where a carrier has not disclosed a detail, we write "Not disclosed / verify with carrier" instead of guessing.
- Visible changelogs. The comparison table and exclusion tracker each publish a dated changelog, because in this market freshness is the whole point.
- Quarterly re-verification. Every cell is re-checked against its source at least quarterly, with ad hoc updates on launches and filings.
- Verification status is explicit. Rows that we have not yet confirmed against a primary source are labeled as needing primary verification rather than presented as settled fact.
How this guide is funded
The editorial work, which rankings and which products are included, how each claim is verified, and what the datasets say, is never for sale. Nothing an advertiser, sponsor, or client does can change the datasets or the analysis.
The work is funded three ways, all disclosed:
- Display advertising. The site runs display ads. Advertisers have no say over research, rankings, or verification.
- Clearly-labeled sponsorships. Any sponsored placement is marked as such and kept separate from the editorial datasets.
- Optional data and advisory services. iSinghLabs offers AI governance and coverage-readiness advisory work. Engaging those services never buys editorial treatment here.
The line that does not move: no carrier pays to appear in this guide, and no commission is earned on any policy. Nothing influences the editorial datasets. The full funding disclosure is on the About page.
Informational only. Nothing in this brief is insurance advice, legal advice, or a recommendation of any policy. Coverage terms, form adoption, and state approvals change frequently and vary by jurisdiction and insured. Verify every detail, including the exact endorsement wording on your own policy, with your carrier or a licensed broker. Research by iSinghLabs Inc; reviewed by iSL Advisory. Author: Joel R. Singh. Edition updated 2026-07-12.