By the Editors · Published 12 June 2026 · Last updated 12 June 2026 · 10 minute read

If your AI agent gives wrong advice, will insurance actually pay out?

Q: If my AI agent gives wrong advice and a customer complains, does my professional indemnity policy respond?

It depends on two things: whether your PI policy was written to include automated digital services, and whether you notified your insurer at policy inception that your business uses AI agents to give advice. Most professional indemnity wordings written before 2024 treat advice as a human act. An AI agent giving advice in your name will fall into coverage only if the policy language is broad enough to cover your firm's services regardless of how they are delivered. If you did not disclose AI use at last renewal, the insurer may argue non-disclosure and reduce or deny the claim.

Your AI agent told a customer the wrong thing. Maybe it quoted the wrong price. Maybe it gave advice about a product the customer was not eligible for. Maybe it made a claim about your service that turned out to be incorrect. The customer is now unhappy and wants something done about it. This is the question every SME operator has in the back of their mind when they deploy an AI agent. The honest answer is: it depends, and the details matter more than you might expect.

Key takeaways

Most professional indemnity and general liability policies written before 2024 were not designed to cover AI agent outputs. Unless your broker specifically asked about AI use at inception, you may have a gap that neither party was aware of at the time.
In 2023, ISO introduced two new AI exclusion endorsements for commercial general liability: CG 40 47 (broad exclusion) and CG 40 48 (limited exclusion). Policies with CG 40 47 exclude most AI-related harm. Policies with neither endorsement may offer broader unintended cover, but renewals are adding these clauses fast.
The British Columbia Civil Resolution Tribunal confirmed in Moffatt v. Air Canada (February 2024) that a business is responsible for representations made by its AI agent. You cannot disclaim the agent as a separate entity. Your liability to the customer is likely real before you even reach the insurance question.
Cyber insurance covers data and network events. It does not cover your liability for wrong advice from an AI agent unless a cyber event caused the error. These are different products covering different risks.
New AI-specific liability products from HSB (Munich Re), Armilla (Lloyd's), and Counterpart now offer affirmative AI errors and omissions coverage. If you deployed an AI agent before buying one of these products, check whether the policy covers incidents that occurred before the policy incepts.

Why the standard policy answer is "probably not"

The British and European commercial insurance market was not built to insure AI agent outputs. The standard professional indemnity policy covers a professional providing advice or services to a client. The standard commercial general liability policy covers bodily injury, property damage, and personal and advertising injury. Neither policy template was written with the concept of an autonomous AI agent in mind.

When an AI agent gives a customer wrong advice and that customer suffers a loss, the resulting claim does not fit neatly into either box. It is not a classic negligence claim against a professional's judgment. It is not a product defect claim against a physical product. It is a claim arising from the automated output of a system the operator deployed, configured, and chose to put in front of the customer. The legal theories that apply are still being worked out in courts. The insurance products being used to respond to those claims were not designed for that purpose.

The gap between what operators need and what existing policies provide is one of the defining insurance market problems of the next three years. Some insurers are reacting by adding AI exclusion endorsements to limit their exposure. Others are building new affirmative AI coverage products. Neither has reached the scale where an SME operator can simply ask their broker "am I covered?" and receive a confident yes. The correct response right now is: review what you have, understand what it actually says, and decide whether you need something different.

The three policies that might respond, and what each actually covers

Professional indemnity (also called errors and omissions)

Professional indemnity insurance is the most likely policy to respond to a claim arising from wrong advice delivered by an AI agent, because it is designed to cover the professional advice and services you provide to clients. The key question is whether your policy language is broad enough to cover advice delivered through an automated system rather than a human professional.

Older PI wordings define the insured activities narrowly: the practice of a named profession, the provision of named services, or the acts of the named insured's employees. If your AI agent does not fit within any of those definitions, the insurer will argue it is not an insured activity. Newer wordings in the technology professional indemnity sector are broader and cover "technology services" or "digital services" delivered by or through automated systems. If your broker arranged cover through a technology PI product, you are more likely to have coverage than if you bought a standard professional indemnity policy designed for a law firm or accountancy practice.

The second obstacle is disclosure. PI policies are written on a claims-made basis and require you to disclose material facts at inception. If you did not tell your insurer at last renewal that your business uses AI agents to provide advice or recommendations to customers, the insurer may have grounds to argue non-disclosure or misrepresentation. This does not automatically mean the claim is denied, but it creates a vulnerability. The insurer will investigate what they knew and what they would have done if they had known. Some insurers would have offered the same terms. Some would have added an exclusion. Some would have declined to insure. Which of those three would have happened in your case is the question your legal counsel will face.

Commercial general liability

Commercial general liability covers bodily injury, property damage, and personal and advertising injury. A wrong-advice claim from an AI agent almost always involves pure economic loss: the customer lost money because they acted on incorrect information. Most CGL policies explicitly exclude pure economic loss in favour of physical harm or property damage. This means that for most AI advice errors, your CGL policy is not the right tool.

The ISO AI exclusion endorsements are particularly relevant here. ISO CG 40 47, introduced in December 2023, is a broad AI exclusion that removes coverage for bodily injury, property damage, and personal and advertising injury "caused directly or indirectly by any artificial intelligence." If your insurer added this endorsement to your CGL at your last renewal, and many did without making the addition prominent in the renewal discussion, your CGL policy now provides no cover for any harm caused by your AI agent in any category.

ISO CG 40 48 is a more limited endorsement that preserves coverage for AI that "does not independently make decisions" and does not "autonomously operate without human oversight." Whether your AI agent meets this definition is a question of fact and degree. An agent that operates fully autonomously, makes booking decisions, or commits your business to terms without human review, fails this test. An agent that presents options for human approval may pass it. The practical difficulty is that most operators cannot articulate precisely how their agent makes decisions and how much oversight they apply, because they set it up and then let it run.

Cyber insurance

Cyber insurance covers three main categories of loss: first-party costs arising from a data breach or network security event (forensics, notification, business interruption), third-party liability for data breach claims from affected individuals or regulators, and sometimes ransomware or extortion payments. None of these categories naturally encompasses an AI agent giving a customer wrong advice unless the wrong advice was caused by a cyber security event.

The one scenario where cyber coverage might respond is prompt injection: a deliberate attack in which a malicious actor manipulates your AI agent's context or instruction set to cause it to produce harmful outputs. If an attacker caused your agent to quote wrong prices or give fraudulent instructions and your customer suffered loss as a result, the cyber security event could be the triggering cause, and your cyber policy's network security liability section might respond. However, this requires the attack to be the proximate cause, which is a higher bar than the attacker merely having influenced the output.

Cyber insurers are aware of the confusion between AI liability and cyber liability. Several markets are explicitly excluding AI-agent liability from cyber wordings in 2026 renewals, or adding sublimits. If your renewal is coming up and your business uses AI agents, ask your broker to show you the AI-related language in the cyber renewal explicitly, not just the summary schedule.

The Air Canada case and what it means for your policy position

The most cited AI liability case in insurance discussions is Moffatt v. Air Canada, decided by the British Columbia Civil Resolution Tribunal in February 2024.^[1] Jake Moffatt asked Air Canada's chatbot about its bereavement fare policy and was told he could book full-price tickets, travel, and claim the bereavement discount retrospectively. Air Canada's actual policy required the discount to be applied before travel. He travelled, claimed the discount, and Air Canada refused to pay it on the grounds that the chatbot had given him incorrect information. The tribunal ordered Air Canada to honour the discount and rejected the airline's argument that the chatbot was a separate legal entity whose errors were not the company's responsibility.

The case matters for insurance because it settles the liability question before the insurance question arises. If you run an AI agent that makes representations to customers about your prices, terms, eligibility rules, or service scope, you are responsible for those representations. You cannot use the agent as a liability firewall. The fact that a machine said it rather than a person says it does not change the outcome for the customer, and the tribunal confirmed it should not change the outcome for the operator either.

Once your liability to the customer is established, the insurance question becomes: does any of your coverage respond to that liability? The coverage analysis in the previous section applies. The Moffatt case does not itself help you answer the insurance question, but it removes any possibility that you can avoid the question by arguing the agent was a separate entity.

New products that actually cover AI agent errors

Three carriers are now offering affirmative AI agent liability coverage in the UK and European markets. These products were not available two years ago. They exist because the gap in conventional coverage has been widely documented and there is a market for it.

HSB (Hartford Steam Boiler, a Munich Re subsidiary) launched AI Liability Insurance for SMEs in the UK in March 2026. The product covers third-party claims arising from AI errors, including wrong advice, incorrect decisions, and harmful automated actions. It is structured as a standalone AI errors and omissions policy and does not require existing coverage from the same carrier. Premiums for SMEs start at approximately GBP 1,500 per year for modest AI deployments.^[2]

Armilla, a Lloyd's coverholder using Chaucer as the capacity provider, offers AI Performance Insurance that covers the cost of model errors and the downstream liability they create. Armilla's product line includes AI errors and omissions coverage with limits up to USD 25 million following a January 2026 capacity increase.^[3]

Counterpart, backed by Apollo Global Management, offers affirmative AI coverage through management liability and errors and omissions products specifically designed for technology companies deploying AI. The product explicitly addresses AI agent decisions and autonomous outputs as insured events.^[4]

All three products have underwriting requirements. You will need to disclose what AI systems you use, how they are configured, what the human oversight arrangements are, what incidents have occurred, and whether you have a documented governance process. The more documentation you can provide, the better the terms you will receive. A business that cannot describe its AI governance process is both underinsured and underprotected in the event of a claim.

What you should do before your next renewal

The practical steps are straightforward but require action before your renewal date, not after an incident.

First, identify every place in your business where an AI agent, chatbot, or automated tool interacts with customers, produces outputs that customers rely on, or makes decisions that affect customers. This includes chat interfaces, email automation, recommendation engines, scheduling tools, pricing tools, and document generation. Most operators are surprised by how many AI-adjacent systems they actually run once they start listing them.

Second, locate your current professional indemnity and general liability policies and search for any AI-related language. Look for the words "artificial intelligence," "automated," "machine learning," "AI exclusion," "CG 40 47," or "CG 40 48." If you find an AI exclusion endorsement, note whether it is a broad or limited exclusion and assess whether it applies to your deployment.

Third, before your next renewal, prepare a one-page summary of your AI deployments: what each system does, who oversees it, what the incident history has been, and what governance you have in place. Give this to your broker with explicit instructions to disclose it to your PI and CGL insurers and to request coverage confirmation or a coverage endorsement. If your broker cannot confirm coverage, this is the moment to explore the specialist AI products described above.

The three diagnostic questions on this site are designed to help you structure this process. The Coverage Audit tool produces the document your broker needs. The Agent Certified assessment builds the governance evidence that makes your underwriting submission credible. None of these steps takes more than a working day. They are worth significantly more than a day of disruption if something goes wrong.

Frequently asked questions

If my AI agent gives wrong advice and a customer complains, does my professional indemnity policy respond?

It depends on whether your PI policy was written to include automated digital services, and whether you notified your insurer at inception that your business uses AI agents to give advice. Most professional indemnity wordings written before 2024 treat advice as a human act. An AI agent giving advice in your name will fall into coverage only if the policy language is broad enough to cover your firm's services regardless of how they are delivered. If you did not disclose AI use at last renewal, the insurer may argue non-disclosure and reduce or deny the claim.

What is an AI exclusion endorsement and does my policy have one?

An AI exclusion endorsement is a clause added to a policy that specifically removes coverage for losses arising from the use of artificial intelligence systems. ISO introduced two endorsements in 2023: CG 40 47 (a total AI exclusion for commercial general liability) and CG 40 48 (a limited exclusion preserving some coverage). Many business owners have not been told which endorsement applies to their renewal. Request your policy schedule and search for these references.

My AI agent told a customer the wrong price and they now want the difference. Am I covered?

This scenario sits at the intersection of contract law and insurance. The British Columbia Civil Resolution Tribunal found in Moffatt v. Air Canada (2024) that a business is responsible for the prices and terms its AI agent communicates to customers. Your liability to the customer is likely real. Whether your insurance responds depends on whether your policy covers economic loss claims arising from incorrect information provided in the course of business. General liability policies typically do not cover pure economic loss; errors and omissions or professional indemnity policies are more likely to respond, provided AI use was disclosed and no AI exclusion applies.

Does cyber insurance cover AI agent mistakes?

Cyber insurance covers losses arising from data breaches, network intrusions, and system failures. It does not generally cover liability arising from incorrect advice or wrong information provided by an AI agent, unless that incorrect information was caused by a cyber security event such as a prompt injection attack. Your existing cyber policy probably does not cover your AI advice liability exposure. New specialist AI products from HSB, Armilla, and Counterpart offer this coverage separately.

What should I do right now if an AI agent error has already happened and a customer is complaining?

Act in this order: do not admit liability or offer payment before consulting your broker and legal counsel. Notify your insurance carrier of a potential claim event as soon as possible. Most policies require prompt notification as a condition of coverage, and late notification is the most common reason insurers deny AI claims. Preserve all logs and conversation records. Do not alter or delete anything. Assess whether the incident requires notification to a data protection authority (72 hours under GDPR if personal data was involved). Then work with your legal counsel to understand the contractual position.

References

Moffatt v. Air Canada, 2024 BCCRT 149 (BC Civil Resolution Tribunal, February 14, 2024). Tribunal member Christopher Rivers. The tribunal awarded CAD 812.02 and rejected Air Canada's argument that its chatbot was a separate legal entity for whose statements it bore no responsibility.
HSB (Hartford Steam Boiler Inspection and Insurance Company, a Munich Re subsidiary) UK AI Liability Insurance product launch, March 2026. HSB provides specialist engineering and technology insurance in the UK market. Premium figures based on published indicative pricing for SME deployments as of Q1 2026.
Armilla AI Inc., Series A funding announcement, January 2026. USD 25 million capacity increase. Armilla operates as a Lloyd's coverholder with Chaucer Syndicate as the capacity provider. AI Performance Insurance product covers model errors and downstream liability. See the Get Covered directory for current product information.
Counterpart Inc. AI liability coverage products. Backed by Apollo Global Management. Affirmative AI coverage for management liability and E&O, explicitly addressing autonomous agent outputs as insured events. Product available in US and UK markets as of Q1 2026.
ISO CG 40 47 (Exclusion, Artificial Intelligence) and ISO CG 40 48 (Exclusion, Certain Artificial Intelligence). Commercial general liability endorsements introduced by Insurance Services Office in December 2023. For current endorsement language and advisory notices, consult your broker or ISO's Forms portal.