Skip to content
MAYFLOWER SPECIALTYMayflower Specialty
Get a quote

Coverage module

AI Professional Liability (AI-E&O)

AI-E&O is the professional liability module from Mayflower Specialty, written for claims that a company's AI-enabled product or service caused a client or third party a loss. It is built for exposures such as hallucinated or inaccurate output, model errors and drift, and mistakes by AI agents acting for clients, where technology and professional E&O wordings may be silent on AI or exclude it.

Get an AI-E&O quoteTalk to us

At a glance

Who brings the claim
Clients, customers and other third parties who relied on the AI
Policy form
Claims made and reported
How it is bought
As a primary module, or through AI DIC Excess over the E&O policy
Written on
A- (Excellent) AM Best rated paper

Definition

What Is AI Errors and Omissions Insurance?

Definition

AI errors and omissions insurance is professional liability cover that addresses AI in its wording, so that claims alleging an AI-enabled product or service caused a client a loss are dealt with in the policy rather than argued over after the loss. It is sometimes called AI hallucination insurance, although hallucinated output is only one of the exposures it addresses.

When AI output is wrong, the company that put the AI in front of its clients is usually the one that answers for it, whether the model was built in-house or licensed from a vendor. In Moffatt v. Air Canada a tribunal treated a chatbot's answer as the airline's own statement, and in Mata v. Avianca the lawyers who filed citations invented by ChatGPT were sanctioned for the filing, so a business should expect to be held to its AI's output as it would be to its staff's work.

Underwriters see the same concentration of risk: a Lloyd's Market Association survey of 144 respondents, mostly underwriters, in the second and third quarters of 2025 rated professional indemnity as the class of insurance with the highest potential impact from AI loss scenarios, followed by cyber.

Sources Mata v. Avianca, S.D.N.Y. opinion and order (ECF 54) (opens in a new tab)Lloyd's Market Association, AI loss scenarios survey (2025) (opens in a new tab)

Real case

Decided

February 14th 2024

Moffatt v. Air Canada

A British Columbia tribunal held Air Canada liable for negligent misrepresentation after its website chatbot gave a customer wrong advice about bereavement fares. It rejected the argument that the chatbot was a separate entity responsible for its own actions, observing that it “is still just a part of Air Canada's website,” and ordered C$812.02 in damages, interest and fees.

Source Moffatt v. Air Canada, 2024 BCCRT 149 (opens in a new tab)

Written for

What AI-E&O Is Written For

AI-E&O is written for claims by clients and third parties who suffer a loss from AI-enabled products or services, such as an inaccurate or hallucinated output. These claims usually take one of four forms.

Hallucinated or Inaccurate Output

A generative AI feature produces a confident but wrong answer, a fabricated citation or a misleading summary, and a client acts on it. The claim alleges negligence or misrepresentation in the service, whatever role the model played.

Model Error and Drift

A predictive model that performed well at launch degrades as the data it sees changes, and its pricing, scoring or forecasting errors cause client losses before anyone notices, so these claims usually turn on how the model was monitored and how changes to it were controlled.

Mistakes by AI Agents

An AI agent with authority to act, for instance by sending messages, changing records or moving money, takes the wrong action for a client, and the client looks to the company that deployed the agent to make good the loss.

Reliance on Third-Party Models

Many AI products are built on models licensed from vendors whose contracts may cap what can be recovered from them. When the model fails, the client sues the company it contracted with, which is left to absorb the loss beyond any vendor recovery.

Whether a particular claim is covered depends on the policy wording and the facts. AI-E&O is written for these exposures, subject to the policy terms.

The record

How AI Output Has Become Liability

Each decision and incident below is sourced, and together they show tribunals holding businesses to their AI's output, professionals answering for work they did not check and an AI agent deleting a production database during a code freeze. A company that sells AI-enabled work should review its client contracts and its E&O wording with these cases in mind.

  1. June 22nd 2023

    Sanctioned

    Mata v. Avianca

    A federal judge in New York fined two lawyers and their firm $5,000 after they filed a brief citing six cases that ChatGPT had invented, and ordered them to write to each judge falsely named as an author of the fake opinions.

    Source Mata v. Avianca, S.D.N.Y. opinion and order (ECF 54) (opens in a new tab)

  2. May 19th 2025

    Decided

    Walters v. OpenAI

    A Georgia trial court granted OpenAI summary judgment in the first AI-hallucination defamation case to be decided, finding no defamatory meaning, no fault and no damages. The claim failed, but it still had to be defended through to summary judgment.

    Source Gibson Dunn on Walters v. OpenAI, May 2025 (opens in a new tab)

  3. July 2025

    Incident

    An AI Coding Agent Deletes a Production Database

    During a code freeze, Replit's AI coding agent deleted a user's production database, which the company later restored. The incident shows how quickly an agent with write access can cause a loss for the people relying on it.

    Source Fortune, July 23rd 2025 (opens in a new tab)

2,100+

Court decisions worldwide involving AI-hallucinated content, as of October 2026

Source: Damien Charlotin, AI Hallucination Cases database

88%

Organizations that used AI in at least one business function in 2025, according to the Stanford AI Index 2026

Source: Stanford AI Index 2026

Exposure by role

Users, Providers, Integrators and Developers

The supplemental application asks each applicant how it deploys AI, because the role changes who can bring a claim and what they will allege.

AI deployment roles described in the supplemental application
RoleWhat it meansHow the exposure changes
AI userUses AI for its own internal or operational workClaims arise when AI-assisted work reaches a client, such as a report, a recommendation or advice
AI provider, embeddedBuilds AI into the products or services it sellsCustomers can claim that an AI feature failed to perform as described or caused them a loss
AI provider, standaloneSells AI models, platforms or APIsCustomers build on the product, so one failure can reach many downstream businesses
AI integratorDeploys third-party AI under its own brandClients sue the brand they contracted with, while the model belongs to someone else
AI developer, foundationTrains and releases foundation modelsExposure extends to how others use the model, and claims can come from parties with no contract

The roles are taken from the supplemental application, which asks the applicant to select every role that applies. A company that both sells AI features and uses AI in its own work should expect underwriters to look at both exposures.

Existing cover

Why an Existing E&O Policy May Not Respond

A technology or professional E&O policy may respond to a claim that AI-assisted work caused a client loss, because that is a claim about the service. Cover can fall away at the edges, however, where the policy defines professional services narrowly, excludes AI or leaves it unaddressed.

Narrow Definitions of Professional Services

Many E&O policies cover only the services described in the declarations. If an AI feature or an AI-assisted service sits outside that description, the insurer can argue that the claim falls outside the policy.

AI Exclusions

The “absolute” AI exclusion introduced by W. R. Berkley, as reported by Hunton Andrews Kurth, applies to E&O as well as to D&O and fiduciary liability. The wider market is moving the same way: Insurance Business reported in September 2026 that W. R. Berkley, Chubb, Travelers, Berkshire Hathaway and AIG had filed to adopt the ISO generative AI forms or their own equivalents by April 2026.

Sources Hunton Andrews Kurth, May 2025 (opens in a new tab)Insurance Journal, August 17th 2026 (opens in a new tab)Insurance Business, September 16th 2026 (opens in a new tab)

Contract and Performance Guarantees

E&O forms commonly exclude liability assumed under contract and guarantees of performance. A client contract that promises a level of accuracy for an AI feature can create exposure the policy does not pick up.

Before renewal

Review how the E&O policy defines professional services, and whether it mentions AI, at least 90 days before renewal, and check that the AI features you sell or use in client work fall inside that definition.

Your program

How AI-E&O Fits With Your E&O Policy

AI-E&O can be bought in two ways, depending on what the existing technology or professional E&O policy says about AI.

Option A

As a Primary Module

AI-E&O is a primary module written for client and third-party claims arising from AI-enabled products and services, bought alongside the E&O policy the company already carries. It suits a company whose AI features or AI-assisted services are central to what it sells.

Start an AI-E&O application

Option B

Through AI DIC Excess

Where the existing E&O policy excludes AI or is silent on it, AI DIC Excess sits over the program and is written to respond where the underlying policy does not, subject to its terms.

How AI DIC Excess works

Scenarios

AI-E&O Claim Scenarios

These scenarios show how AI-enabled products, services and agents turn into client claims across different industries.

These scenarios are hypothetical, and whether a policy responds depends on its wording and the facts.

Professional services

A consulting firm delivers a market study drafted with generative AI. The client relies on it to enter a new market, then discovers that several of the cited sources do not exist and sues for the cost of the failed launch.

Hypothetical · Written for: AI-E&O

Software

A software company's AI contract-review feature misses automatic renewal clauses in hundreds of customer contracts. A customer claims for the cost of renewals it could no longer cancel.

Hypothetical · Written for: AI-E&O

Payments

A payments company's AI agent, authorized to issue refunds, misreads an instruction and issues duplicate refunds across thousands of a merchant client's accounts. The client sues for the loss and the cost of recovering the payments.

Hypothetical · Written for: AI-E&O

Who it is for

Who Needs AI-E&O

AI-E&O is designed for companies whose AI-enabled products, services or agents could cause a client or third party a loss.

  • Companies that build AI into the products or services they sell, or sell AI models, platforms and APIs
  • Professional services firms that use AI in client work, such as research, drafting, analysis or advice
  • Companies whose customer-facing chatbots or assistants give people information they act on
  • Companies that let AI agents act for clients, for example by sending messages, changing records or moving money
  • Companies whose E&O renewal has added an AI exclusion, which may also want to consider AI DIC Excess

Where AI-E&O Is Not the Answer

  • Employment claims over AI hiring and workforce tools, which AI-EPL is written for
  • Securities and oversight claims against the board over AI, which AI-D&O is written for
  • First-party security losses such as breach response, ransomware or business interruption, which belong with cyber insurance

The supplemental application also places these uses of AI outside Mayflower's underwriting appetite:

  • Autonomous weapons or military targeting
  • Social scoring or mass surveillance
  • Biometric identification in public spaces
  • Real-time emotion inference in workplace or educational settings
  • Deepfake generation without disclosure

Underwriting

What Underwriters Will Ask

Mayflower underwrites on the applicant's AI governance, so the questions that matter most for AI-E&O concern the guardrails around AI output, how models are monitored and what happens when a vendor's component fails. The questions below are plain-language summaries of the supplemental application, which has the exact wording.

Start the applicationDownload the application

Completing the application does not bind coverage. How underwriters assess AI risk

  1. Section II · AI systems overview

    Which guardrails apply to generative AI in production, such as grounded retrieval with source citations, output moderation, hallucination detection, logging and human review before output reaches a customer?

    Why it is asked: These controls decide how often a wrong answer reaches a client.

  2. Section II · AI systems overview

    Do any AI agents in production have write access or financial authority, and if so, what transaction limits, approval gates and audit logs apply?

    Why it is asked: Agents that can act carry a different exposure from tools that only advise.

  3. Section II · AI systems overview

    How much does the company rely on third-party AI components, and do the contracts include support terms, indemnities and a documented fallback plan?

    Why it is asked: Vendor terms decide how much of a loss can be recovered from the model provider.

  4. Section V · System operations and monitoring

    Is performance and drift monitoring automated, with alerts to a named owner, or reviewed periodically?

    Why it is asked: The speed of detection limits the size of a drift claim.

  5. Section V · System operations and monitoring

    Are new model versions tested in shadow or A/B deployments before full rollout?

    Why it is asked: Staged releases catch regressions before every client sees them.

  6. Section III · AI governance

    Which reviews are required before a new AI system goes live, such as a signed risk assessment, red-team testing, legal review and a security review?

    Why it is asked: Pre-deployment review shows that the company checked the system before clients relied on it.

  7. Section VI · AI incident response

    Does the company have an AI-specific incident response plan that has been tested in the past 12 months?

    Why it is asked: A tested plan shortens the time between an AI error and its correction.

FAQ

Questions About AI-E&O

Does E&O insurance cover AI mistakes?

A technology or professional E&O policy may cover a claim that AI-assisted work caused a client loss, because the claim is about the service. Cover depends on whether the AI work falls within the policy's definition of professional services and on whether the policy excludes AI, so the wording should be checked. AI-E&O is written to address these claims affirmatively, subject to the policy terms.

Can you insure against AI hallucinations?

Liability arising from hallucinated output can be insured, because the resulting claims are professional liability claims alleging that a product or service caused a loss. AI-E&O is written for claims by clients and third parties who suffer a loss from inaccurate or hallucinated output, subject to the policy terms, and underwriters look closely at the guardrails a company uses around generative AI.

Is a company liable for what its chatbot tells customers?

A company can be liable for what its chatbot tells customers. In Moffatt v. Air Canada, decided on February 14th 2024, a British Columbia tribunal held the airline liable for negligent misrepresentation after its website chatbot gave a customer wrong fare advice, and rejected the argument that the chatbot was responsible for its own actions.

Does AI-E&O respond when the faulty AI came from a vendor?

Claims usually follow the company that delivered the product or service to the client, even when the model came from a vendor, and vendor contracts may cap what can be recovered from the vendor. AI-E&O is written for claims arising from a company's AI-enabled products and services, including those built on third-party models, subject to the policy terms.

Which guardrails do underwriters look for?

Underwriters look for input validation, retrieval that grounds answers in cited sources, output moderation, hallucination detection, logging of prompts and outputs, red-team testing and human review before output reaches a customer. For agents, they also look at transaction limits, approval gates and audit logs, and for vendor models, at contracts, indemnities and fallback plans.

Further reading

Guides and Related Coverage

Other modules

AI-E&O

Stand Behind the Work Your AI Delivers

Apply online, or talk to the team about how AI-E&O would sit with your current professional liability cover.

Get a quoteTalk to us

Placed through brokers on A- (Excellent) AM Best rated paper.