To find out whether your policy excludes artificial intelligence, open three places in it: the endorsements schedule, the definitions section and the exclusions section, and compare each with last year’s wording. The 2026 renewal season brought written AI exclusions into general liability and management liability policies, while part of the London market started writing affirmative AI cover into technology professional liability (Tech E&O) and cyber policies.
This page is written for management and in-house counsel at Israeli technology, SaaS and AI companies approaching renewal. Policies that stay silent on AI are covered in Gray Areas: When AI Failures Aren’t Covered by Insurance. Here the policy already speaks about AI explicitly, and the question is what it says and how to change it before you sign.
Key points
The 2026 exclusions are written down – Since January 2026 three standard US endorsements that exclude generative AI from commercial general liability have been available: CG 40 47, CG 40 48 and CG 35 08. By the end of July 2026 about 2,370 filings for such exclusions were already in force across 49 states.
An “absolute” exclusion goes much further than a generative-AI exclusion – Absolute wordings in D&O and professional liability policies remove any claim “arising out of” the use of AI by any person, including the company’s own statements about its AI.
Part of the market is moving the other way – In the London market, including at Lloyd’s, some insurers have added wording to Tech E&O and cyber policies confirming that AI involvement does not stop the policy from responding, while others have placed AI events under a sub-limit.
The wording in force when the claim is made decides – Under a claims-made policy, an exclusion added at the 2026 renewal may apply to a model error from 2024, which is why the exclusion has to be negotiated before you sign.
What changed in 2026: from silent AI to written exclusions
Until 2025 most commercial policies did not mention artificial intelligence, and whether they responded to an AI event was left to the claim stage. In 2026 the market began answering that question in writing, mostly by narrowing cover.
The clearest signal is the set of standard endorsements from ISO (Insurance Services Office), the body that drafts standard policy forms for the US market. CG 40 47 removes bodily injury, property damage and personal and advertising injury arising out of generative AI from the CGL policy. CG 40 48 removes personal and advertising injury only (Coverage B). CG 35 08 applies the exclusion to the products-completed operations coverage part. State filing data showed more than 4,000 records by the end of July 2026, about 2,370 of them already in force.
A different kind of exclusion appeared in management liability at the same time. In 2025 a US specialty insurer introduced an “absolute” AI exclusion for its management and professional lines, and similar wordings have since appeared elsewhere.
London has taken a different path. As of late October 2026 no uniform AI exclusion clause had been published in the Lloyd’s market. Some specialist London insurers updated their Tech E&O and cyber wordings so that AI involvement does not prevent the policy from responding, while others introduced AI sub-limits, sometimes at around a tenth of the policy limit.
The result: two policies at similar prices may respond in opposite ways to the same claim.
Three families of exclusion, and what each one removes
Not every AI exclusion does the same damage. The difference between an absolute wording and one limited to generative AI or to content can separate a policy that responds to most claims against a software company from one that barely responds at all.
The absolute AI exclusion
The absolute wording removes any claim “based upon, arising out of, or attributable to” the use or development of artificial intelligence by any person or entity. Its list includes products that incorporate AI, promises made by a chatbot, the company’s statements about its AI and regulatory demands to assess AI risks. For a company whose product is built on AI, a wording like this can empty the policy.
The generative AI exclusion
The standard US endorsements are limited to generative AI and carry their own definition of the term. They attach to the general liability policy and the products coverage part, so they bite mainly on bodily injury, property damage and advertising claims. A causal link is enough: if a generative tool was used at one stage of the work, the insurer may argue that the exclusion reaches the whole claim.
The content and advertising exclusion
The narrowest wording removes only personal and advertising injury arising from generative content, such as defamation in marketing material. For a company that publishes large volumes of content with generative tools, this is still a real exposure.
| Exclusion type | What it removes | Where it usually appears | What to ask for at renewal |
| Absolute AI exclusion | Any claim connected to the use or development of AI, including statements about AI | D&O and professional liability | Deletion, or a narrow wording with carve-backs for defence costs and Side A |
| Broad generative AI (CG 40 47) | Bodily injury, property damage and advertising injury from generative AI | US general liability | Check the definition of “generative AI” and the causal test |
| Content and advertising only (CG 40 48) | Personal and advertising injury from generative content | General liability and media policies | Confirm the rest of the policy stays clean |
| Products-completed operations (CG 35 08) | Bodily injury and property damage from a product with a generative AI component | Products coverage part | Review the product liability policy |
| AI sub-limit | The gap between the sub-limit and the actual loss | Cyber and Tech E&O | Compare the sub-limit with a realistic worst case |
| Affirmative AI endorsement | Any scenario not listed, if it is drafted as a closed list | Tech E&O and cyber | A wording stating that AI involvement does not bar cover |
Where the exclusion hides in the policy
An AI exclusion almost never appears on the declarations page. It usually sits in the endorsements schedule or in the definitions, and sometimes inside an old exclusion that has been reworded.
The endorsements schedule and the definitions
Start with the form numbers: a side-by-side list of this year’s endorsements against last year’s, and a full reading of every new form. Then move to the definition, which sets the boundaries of the exclusion. Some 2026 wordings copy the EU AI Act definition: a machine-based system that infers from the input it receives how to generate outputs such as predictions, content or decisions. A definition like that can catch a recommendation engine or a code-completion tool, even at a company that does not see itself as an AI business, and certainly at one that runs autonomous AI agents.
Old exclusions and small words
Not every exclusion is labelled “AI”. In a Tech E&O policy, check whether an exclusion has been added for “automated decisions” or for “outputs not reviewed by a human”. Then check the causal language. “Arising out of” is the broadest phrase: it is enough for AI to be one link in the chain. “Directly caused by” is narrower and a reasonable opening position in negotiation.
The proposal form and excess layers
2026 questionnaires ask which AI tools the company uses and who reviews the outputs. Under section 6 of Israel’s Insurance Contract Law, 1981, the insured must give full and honest answers to questions the insurer put in writing, and an inaccurate answer may allow the insurer to reduce the indemnity, subject to the law. The certificate your customer receives shows the line and the limit, not the exclusions, so an enterprise customer may only learn about the exclusion at claim stage. An excess policy written on a follow-form basis inherits the exclusion, and an excess insurer may add its own exclusion even above a clean primary layer.
Line by line: what the exclusion may remove
Each line of insurance was built to respond to a different kind of claim, so an AI exclusion hurts each one in a different place, subject to the specific wording.
| Line of insurance | What it is designed to respond to | What an AI exclusion looks like in it | The limitation to know |
| Technology professional liability (Tech E&O) | A customer’s financial loss from a software error | Exclusion of AI outputs or of products that incorporate AI | For an AI-based product it strikes at the core of cover |
| Cyber | A security incident and the cost of responding to it | Usually a sub-limit or a clarification | A model error without a breach usually belongs to Tech E&O |
| Directors and officers (D&O) | Claims against officers over decisions and statements | An absolute exclusion that includes statements about AI | AI-washing claims may fall straight into it |
| General liability | Third-party bodily injury, property damage and advertising injury | The standard generative AI endorsements | Not designed for pure financial loss |
If the company sells an AI-based product, ask the insurer to confirm in writing that the product, as described in the proposal form, falls within the definition of professional services. In D&O, check whether the exclusion also applies to Side A, the personal cover for officers when the company cannot indemnify them. The split between the lines is mapped in Tech E&O, Cyber, D&O and Crime: Where Software Risk Meets Insurance.
When one model is wrong a thousand times: aggregation
Professional liability policies were built around human error, one mistake at a time. A model that is wrong is wrong in the same way for every customer, and a single fault can produce dozens of claims at once. The related claims clause decides whether all of them are one claim with one limit and one retention, or separate claims. The answer cuts both ways: one claim may exhaust the limit, while separate claims may each carry a retention. This clause and the AI exclusion together decide how much of the exposure stays with the company.
Four legal systems: how local context changes the picture
The exclusions were born in the US market, but an Israeli company meets them through policies written in London and the US. The law in each region affects how the exclusion will be read and what exposure it leaves.
Israel
Israel has no dedicated AI statute yet, and the Privacy Protection Authority published a draft guideline in 2025 on how the Privacy Protection Law applies to AI systems. In February 2025 the Supreme Court held, in a petition that cited authorities which did not exist, that using AI tools does not release a professional from the duty to check the output. The direction is clear: “the AI got it wrong” does not move liability away from the company, so the question shifts to the policy. Israeli courts tend to read exclusions narrowly and place the burden of proving them on the insurer. A well-drafted exclusion leaves little room for that.
United States
The ISO endorsements are optional. An admitted insurer files them with each state regulator, while surplus lines insurers usually draft their own. Many courts read “arising out of” broadly. Under a policy with a duty to defend, check whether the exclusion also removes defence funding.
European Union
Transparency obligations under Article 50 of the AI Act apply from 2 August 2026. Regulation (EU) 2026/1744, in force since 27 July 2026, postponed the high-risk obligations under Annex III to 2 December 2027 and those under Annex I to 2 August 2028. The new Product Liability Directive (EU) 2024/2853 brings software within the definition of a product and applies to products placed on the market from 9 December 2026. Whether an AI Act fine is insurable depends on each member state’s law, and it should not be assumed to be covered.
Asia-Pacific
South Korea’s AI Basic Act took effect on 22 January 2026, and Japan passed an AI promotion act in 2025 that relies mainly on guidance. India and China generally require a local insurer, so check that the local policy in a global programme does not add an AI exclusion that the master policy does not contain.
Before renewal: what to ask your broker and what can be achieved
The exclusion is managed in the months before renewal. Insurers are willing to discuss wording when the company presents an orderly picture of its AI use and the controls around it.
Start 90 to 120 days before expiry and ask for draft wordings for all lines together, including excess layers. The broker (LAMDA Broking) and legal counsel review the renewal wordings against the insurance clauses in customer contracts, to make sure a new exclusion does not put the company in breach of a contractual commitment.
| What to ask for | When it is realistic | What to show the underwriter |
| Deletion of the exclusion | When AI is marginal to the business, or when there is competition for the risk | A map showing limited internal use of AI tools |
| A narrow wording instead of an absolute one | In most cases, especially in D&O | A separation between AI products sold and internal tools |
| Carve-backs for defence costs and Side A | When the insurer insists on an exclusion | An approval process for AI statements to investors |
| An exclusion that applies only to acts after a set date | When the exclusion is introduced for the first time | The date from which the new systems went live |
| An affirmative AI endorsement or a dedicated AI policy | For AI-based products, in specialist markets and at Lloyd’s | Human oversight of outputs and liability caps in customer contracts |
The pricing of an extension, or of deleting an exclusion, depends on the scale of AI use, the business activity and the US exposure. What a dedicated AI policy may cover is set out in What Does AI Insurance Cover?, and the wider picture of AI, cyber and D&O is in AI, Cyber Insurance, and D&O.
Common mistakes
Most of the gaps that surface in AI claims were created at renewal:
- Reading only the declarations page: the exclusion almost always sits in the endorsements or the definitions.
- Assuming the exclusion is irrelevant because the company does not develop AI: a broad definition also catches code-completion tools and service chatbots.
- Checking one line only: exclusions stack, and the gap opens between policies and between layers.
- Welcoming an affirmative endorsement without reading it: a closed list of covered scenarios may leave every other scenario outside.
Frequently Asked Questions
How do we know whether our policy contains an AI exclusion?
Compare the endorsements schedule with last year’s and search the definitions for terms such as Artificial Intelligence and Generative. Ask your broker for a table of every AI reference across all policies, including the excess layers.
We don’t develop AI, we only use Copilot and ChatGPT. Is the exclusion relevant to us?
It may be. An absolute exclusion applies to the use of AI by any person, not only to development. Code written with a completion tool or marketing copy produced by a language model may fall within it, subject to the wording and the definition in the policy.
What is the difference between an absolute exclusion and a generative AI exclusion?
An absolute exclusion removes any claim arising from the use of AI, including statements about AI, and appears mainly in D&O and professional liability. A generative AI exclusion is limited to model-generated content and appears mainly in general liability. For a company selling an AI product, the difference may decide whether the policy has any value.
Will our cyber policy respond to an AI-related incident?
It depends on the wording. The trend in cyber is more moderate: some insurers confirm that an AI-assisted attack, or downtime caused by a model, triggers the usual cover, while others impose a sub-limit. A model error that causes a customer financial loss without a breach usually falls to Tech E&O.
Can an AI exclusion be removed at renewal?
Sometimes. Some insurers will delete the exclusion, replace an absolute wording with a narrow one, give back cover for defence costs and Side A, or set a sub-limit instead. The chances improve when the company presents an orderly map of its AI systems and starts negotiating several months before expiry.
The exclusion was only added this year. Does it apply to an error from two years ago?
Under a claims-made policy, usually yes. The wording in force when the claim is made decides, so an exclusion added in 2026 may apply to a claim arising from an act in 2024. The fix lies in negotiation, for example an exclusion that applies only to acts after a set date.
Our US customer requires Tech E&O without AI exclusions. What do we do?
First check what the contract actually requires: usually a line of cover and a limit, less often specific exclusions. If the contract prohibits exclusions that narrow cover for the product sold, a broad AI exclusion may create a breach. In that case, ask the insurer for a narrow wording or an affirmative endorsement.
An AI exclusion signed at the 2026 renewal will decide the fate of claims made in the years that follow, and its wording can only be changed before you sign.
The above is general information only and does not constitute insurance, legal or other professional advice. Coverage terms, exclusions and obligations are set by the specific policy wording and are subject to applicable law. Each case should be reviewed on its own merits against the policy wording and with a qualified adviser.






















































































































































