By the Editors · Published 14 June 2026 · Last updated 14 June 2026 · 10 minute read

Does professional indemnity insurance cover AI tools I use in client work?

If you are a consultant, accountant, lawyer, financial adviser, architect, or any other professional who uses AI tools to produce work for clients, your professional indemnity insurance may not protect you when that AI output turns out to be wrong. The policy language was written before AI became standard practice. The question of whether your PI cover responds to an AI-assisted error is live, contested, and getting worse at renewal. This is what to check, what to ask your broker, and what to do if the answer is not reassuring.

What professional indemnity insurance actually covers

Professional indemnity insurance is designed to respond when a professional's advice, service, or work product causes a financial loss to a client, and the client claims that the loss arose from the professional's negligence, error, or omission. The insuring clause in a standard PI policy covers amounts the insured becomes legally liable to pay, and defence costs, arising from a negligent act, error, or omission in the performance of professional services.

Three words in that insuring clause matter for AI: "professional services." PI policies are not product liability policies. They are not technology errors and omissions policies. They respond to human professional failures, not to the failure of a tool the professional used. Whether the tool happened to be AI, spreadsheet software, or a faulty reference work, the claim against the professional is the same: you gave negligent advice or produced defective work product.

This creates a structural question when AI is involved. The tool produced an error. You incorporated the error into work product without catching it. The client lost money. Is the claim covered? The answer is not straightforward, and it depends on several factors that most PI policies do not address explicitly because they were drafted before AI became a standard professional tool.^[1]

The verification question: why what you did with the output matters

The single most important determinant of whether your PI policy responds to an AI-related claim is what you did with the AI output before you delivered it to the client. This is the verification question, and it runs through every professional liability analysis involving AI.

If you used AI to generate a first draft and then reviewed it against independent sources before incorporating it into a final deliverable, the claim would be that your review process was negligent. This is a standard professional negligence analysis: did you take the care that a reasonably competent professional in your position would take? The AI is just the tool. The standard of care is the question.

If you used AI to generate content and delivered it to the client with minimal or no independent verification, the claim has a different character. You relied on a tool you knew, or should have known, produced unreliable outputs in the relevant category without adequate checking. Whether a PI policy responds to this depends on the policy's approach to reckless or deliberate risk-taking. Some policies exclude or limit claims where the insured took undue risks or disregarded known dangers. An insurer who can show that AI hallucination in the relevant context was a widely documented and publicly known risk at the time you used it may argue that your failure to verify was not an error but a knowing acceptance of risk.^[2]

The Mata v. Avianca case is the clearest example of this dynamic. The lawyers knew they had used an AI tool. They submitted the output without verifying the citations. When challenged, they first insisted the citations were real. The court found that their use of AI without verification, followed by their initial insistence that the AI-generated citations were accurate, crossed from negligence into professional misconduct. The sanctions imposed were for the failure to check, not for the AI error itself.^[2]

AI exclusions: what to look for in your current policy

Insurance companies began adding AI exclusions to commercial policies in 2023. The Insurance Services Office introduced exclusion endorsements CG 40 47 and CG 40 48 for general liability policies, addressing exclusions for losses arising from AI systems. These exclusions are now appearing in professional indemnity policy wordings, either as standalone exclusions or as part of broader technology exclusion language.

The most common formulations to look for in your PI policy wording are the following. An exclusion for claims arising from the use of automated decision-making systems, algorithmic tools, or machine learning outputs. An exclusion or sublimit for claims where the professional relied on AI-generated content without independent verification of accuracy. A requirement that the insured confirm their AI use in completing a warranty section of the proposal form, where misrepresentation voids the policy. A carve-out from the professional services definition that excludes activities performed by or substantially delegated to automated systems.^[3]

None of these exclusions are standard across the market yet. Different insurers have taken different approaches and the market is moving rapidly. This is why the explicit broker conversation at renewal is essential: you cannot know your position by reading a generic policy summary.

The disclosure obligation at renewal

Professional indemnity insurance is a contract of utmost good faith. In most European jurisdictions and under the UK Insurance Act 2015, policyholders have a duty to disclose all material facts to their insurer before and at renewal. A material fact is one that would influence a prudent insurer's decision to offer cover, on what terms, and at what premium.

The widespread use of AI tools in professional services is now a material fact for many professionals. If you have materially changed your working practices since your last renewal by introducing AI tools into client deliverables, and you have not disclosed this to your insurer, you carry a non-disclosure risk. If a claim arises that relates to AI use and the insurer discovers you were using AI tools they did not know about, they may decline the claim or void the policy for material non-disclosure.

The disclosure obligation does not require you to volunteer information about every tool you use. It requires disclosure of changes that a reasonable insurer in your sector would consider material to the risk they are underwriting. As AI tool use in professional services has become a subject of specific underwriting attention, there is a reasonable argument that it meets the materiality threshold for most professional services insurers in 2026.^[4]

The practical step is simple: at your next renewal, tell your broker that you use AI tools in client work. Let the broker communicate this to the insurer and get a written confirmation of whether and how this affects your cover. If the insurer declines to confirm that AI-assisted work is within scope, that is important information that should prompt a specific endorsement or a policy change.

What to ask your broker: five specific questions

A general question about "AI coverage" will produce a general answer that may not help you. These five specific questions are more likely to get you the information you need.

First: Does the current policy wording contain any exclusion, sublimit, or condition that applies specifically to AI-generated content, automated systems, or machine learning tools? Ask the broker to check the actual policy wording, not just the product summary. Exclusions are not always flagged in summary documents.

Second: If I use an AI tool to produce a first draft of a report or analysis, and then review and edit that draft before delivering it to a client, and the client later claims the analysis was wrong, would my policy respond to that claim? This is a specific factual scenario. A specific answer is more useful than a general one.

Third: Has the insurer added any new AI-related language to the renewal policy that was not in the prior year's wording? Changes at renewal are sometimes presented as minor updates. Ask specifically whether AI is addressed differently in the new wording.

Fourth: Is there a warranty in the proposal form that requires me to describe how I use AI tools? Some insurers include proposal form questions about automated tools. An inaccurate answer to a proposal form warranty can void the policy.

Fifth: Is there a specialist AI liability policy or endorsement that would provide cleaner coverage for AI-assisted professional work? The market for AI-specific professional liability coverage is developing. Carriers including Counterpart, AIUC, and Lloyd's market syndicates are writing affirmative AI coverage that specifically addresses professional liability for AI-assisted work. Your broker may not raise this unless you ask.^[6]

Protecting yourself in practice: the verification file

The most effective practical protection is not more insurance. It is a verification record that demonstrates you exercised reasonable professional care when using AI tools. This serves two purposes: it reduces the chance that a claim arises in the first place, and it provides evidence that reduces your exposure if a claim does arise.

A basic AI verification file for professional services work contains four elements. First, a record of which AI tools you used for which categories of work, and the AI model version at the time. This establishes what you were working with and protects you if the AI provider later issues a correction about known failures in that model version.

Second, a record of what verification steps you took for each deliverable where AI contributed to a material element. This does not need to be a lengthy document: a timestamped note that you cross-checked the AI-generated analysis against named sources before delivering to the client is sufficient for most purposes.

Third, a client disclosure record. If you disclosed to the client that AI tools were used in the engagement, and the client acknowledged that disclosure, you have established that the client was aware of the methodology and accepted it. This changes the negligence analysis significantly in most professional liability contexts.

Fourth, a note of any AI outputs you reviewed and rejected before including the verified content in the final deliverable. This demonstrates active professional oversight, not passive reliance on AI output.

For the broader question of what documentation you need before an AI agent incident occurs, see the documentation guide for SME operators. For how to read and compare your current policy against AI-specific coverage, see how to choose an AI insurance product.

When PI coverage alone is not enough

For most professional services businesses using AI tools in conventional ways, an updated PI policy with explicit AI coverage confirmation and a strong verification practice will be sufficient. There are three situations where PI coverage alone is structurally inadequate.

First, if you have built or deployed an AI agent that operates autonomously, rather than an AI tool that you use as an assistant, your PI policy is unlikely to cover the agent's independent actions. An AI agent that sends emails, books appointments, processes applications, or takes consequential actions without your direct supervision at each step is not performing professional services under your oversight. It is operating as an extension of your business with its own action-taking capacity. PI insurance covers your professional services. Tech E&O or AI-specific liability coverage is the right instrument for agent operations.^[7]

Second, if you have incorporated AI into a product or service that you sell to clients at scale, rather than using AI as a tool in bespoke professional work, product liability and technology E&O coverage is more appropriate than PI. A practice management platform that uses AI to produce automated risk assessments sold to multiple clients is a product, not a professional service delivered case by case. The EU Product Liability Directive (Directive 2024/2853), applicable from December 2026, treats software including AI components as a product for strict liability purposes.^[8]

Third, if you operate in a regulated profession where sector-specific rules govern your use of AI, the compliance risk runs in parallel to the PI risk. Financial advisers, medical practitioners, and legal professionals in regulated European markets may face specific AI governance obligations from their professional regulators, and a claim arising from breach of those obligations may not be treated as a standard negligence claim under a PI policy. For context on how EU AI Act obligations interact with professional liability exposure, see the EU AI Act operator obligations guide at agentliability.eu.

Frequently asked questions

Related reading

• Who is liable when an AI agent makes a mistake? The 2026 decision tree for operators. Four parties, six legal theories, and the 48-hour incident playbook.
• Does my business insurance cover AI errors? The 2026 policy-by-policy guide. Seven policy types, ISO exclusion endorsements, and named carriers on both sides of the market.
• Documenting AI decisions for insurance purposes: the SME guide. What to record, when, and how to organise it before a claim arises.
• How to choose an AI insurance product: the SME comparison guide. PI versus tech E&O versus dedicated AI liability coverage explained.
• EU AI Act operator obligations: the 2026 compliance guide. The regulatory obligations that run alongside your commercial liability exposure in the EU.

Footnotes

1. For the general structure of PI insuring clauses and their application to technology-assisted professional work, see Lloyd's of London Market Association Professional Indemnity Model Clauses (PI 2019), Section 1. The standard insuring clause covers negligent acts, errors, or omissions of the insured in the performance of professional services and does not define professional services to exclude AI-assisted work.
2. Mata v. Avianca Inc., Case No. 22-cv-01461 (PKC) (SDNY). Order re: sanctions, June 22, 2023 (Judge Kevin P. Castel). Sanctions imposed on Robert Mata, Steven Schwartz, and their supervising partner for submitting AI-generated fabricated case citations without verification.
3. Insurance Services Office (ISO) AI exclusion endorsements CG 40 47 and CG 40 48, introduced in 2023 for commercial general liability policies. The endorsements exclude bodily injury, property damage, and personal and advertising injury arising from the use of AI systems. Equivalent language is appearing in professional indemnity renewal wordings in the UK and European markets.
4. UK Insurance Act 2015, Section 3 (duty of fair presentation). The insured must disclose every material circumstance which the insured knows or ought to know. Under Section 7, a circumstance is material if it would influence the judgment of a prudent insurer in determining whether to take the risk and, if so, on what terms.
5. For guidance on AI tool disclosure at renewal and endorsement language, see the Association of British Insurers (ABI) AI in Insurance guidance (2024) and the Chartered Insurance Institute guidance on AI and professional liability (2025).
6. Counterpart Insurance launched affirmative AI Coverage for Miscellaneous Professional Liability in November 2025. AIUC (Artificial Intelligence Underwriting Company) offers coverage linked to the AIUC-1 certification standard, including professional liability components. Lloyd's of London market syndicates, including those backing Armilla AI (Lloyd's coverholder with Chaucer), are writing AI-specific professional liability endorsements. For current carrier information, see the Get Covered directory.
7. For the distinction between PI coverage for professional services and tech E&O or AI-specific liability coverage for autonomous AI agent operations, see who pays when an AI agent makes a mistake.
8. Directive 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products (revised Product Liability Directive). Article 4(1) explicitly includes software in the definition of product, and Article 4(5) includes digital manufacturing files and AI systems integrated into products within scope. The Directive applies from 9 December 2026.