Does business interruption insurance cover AI agent downtime?
Your AI booking agent goes down for six hours on a Saturday. Orders stop. The phone starts ringing because customers cannot get through any other way. By Monday you have lost a weekend of revenue and paid two staff overtime to cover the gap manually. You call your broker expecting your business interruption policy to help. In most cases it will not, and the reason is a single clause most policyholders have never read closely: the physical damage trigger. This article explains why standard business interruption insurance was not built for software failure, what does actually respond to an AI agent outage, and what to check in your own policy before the next outage happens rather than after.
Key takeaways
- Standard commercial property and business interruption insurance pays for lost income only following direct physical loss or damage to insured property. An AI agent outage caused by a model provider failure, a bad deployment, or a software bug does not damage anything physical, so it typically falls outside the trigger entirely.
- COVID-19 business interruption litigation tested this exact boundary at scale between 2020 and 2022. Courts in the UK, US, and elsewhere largely confirmed that non-physical loss of use, including loss caused by an intangible event with no physical damage, does not satisfy a standard property damage trigger unless the policy contains a specific non-physical extension.
- The coverage that does respond sits in cyber insurance, not property insurance: a system failure or non-physical business interruption extension written into a cyber policy, triggered by an outage of a computer system rather than physical damage.
- Contingent business interruption, which covers loss caused by damage at a supplier's premises, was not designed for a cloud API outage at your model provider and will not respond on standard wording. A small number of cyber policies now extend dependent business interruption to named cloud and API dependencies.
- A handful of dedicated AI performance products, including Munich Re's aiSure, settle against a measured specification tied directly to downtime or performance shortfall against an agreed specification, avoiding the physical damage question entirely.
Why the question matters more than most operators assume
AI agents have moved from experimental to load-bearing in a short period. A business that deployed a chatbot in 2023 as a support supplement may now run its entire first-line sales, booking, or order-taking function through an agent. When that function goes down, the business does not lose a convenience feature. It loses a revenue channel, sometimes the primary one. The financial exposure is not hypothetical: the ElevenLabs decision in February 2026 to insure its AI voice agents explicitly as "insured like any other employee" reflects a market recognition that an AI agent's operational failure is now a material business risk category in its own right, not a minor IT inconvenience.[1]
Yet when operators think about AI agent insurance, they usually think first about liability: what happens if the agent gives a customer wrong information, or makes a decision that harms someone. Business interruption is a different and separate question: what happens to your own income when the agent simply stops working, with no third party harmed at all. Both exposures are real. This article is about the second one, because it is the one most SME operators have not checked and most likely to surprise them at the worst possible time.
The physical damage trigger, explained without the jargon
A standard business interruption policy is not a standalone product. It is almost always an extension attached to a commercial property policy, and it inherits that policy's core logic: the insurer agreed to cover physical assets against physical perils, such as fire, storm, flood, and escape of water. Business interruption cover then pays for the income you lose while those physical assets are being repaired or replaced. The critical structural feature is that business interruption cover is triggered by the same physical damage event that triggers the underlying property claim. No physical damage, no business interruption payment. This is called the physical damage trigger, and it is the default position in the vast majority of commercial property policies sold to SMEs.
An AI agent outage does not involve physical damage to anything. If your booking agent fails because your model provider's API had an outage, because a bad deployment introduced a bug, or because a third-party integration broke, nothing physical has been damaged. Your servers, if you have any, are physically intact. Your premises are untouched. Under the physical damage trigger, there is simply no covered event for the business interruption extension to attach to, regardless of how much income you actually lost.
This is not a novel or AI-specific gap. It is the same structural gap that produced the largest wave of business interruption litigation in insurance history.
The lesson from COVID-19 business interruption litigation
Between 2020 and 2022, courts across multiple jurisdictions heard an enormous volume of business interruption claims from policyholders whose businesses were closed or disrupted by pandemic restrictions, with no physical damage to their premises. The central legal question in nearly every one of these cases was whether a non-physical event, government-ordered closure in that instance, could trigger a policy written on a physical damage basis. In the UK, the Financial Conduct Authority brought a test case that reached the Supreme Court, which found that specific non-damage extensions in certain policy wordings could respond to the pandemic, but only where the policy contained language broad enough to cover non-physical interruption. Policies without such an extension, relying on the standard physical damage trigger alone, were confirmed not to respond.[2]
The direct lesson for AI agent operators is structural, not epidemiological. The pandemic litigation established, at enormous cost and delay to thousands of businesses, that a standard property-based business interruption policy does not stretch to cover a loss of income caused by something other than physical damage, no matter how severe or how directly it affects revenue. An AI agent outage sits in exactly that category: severe, directly revenue-affecting, and entirely non-physical. Insurers who fought and largely won the physical damage argument during the pandemic are not going to concede it for a chatbot outage. Operators who assume their existing business interruption cover extends to AI downtime are relying on an assumption the insurance industry has already litigated against, repeatedly, in a different context with the same underlying legal question.
What does actually cover an AI agent outage
Cyber system failure extensions
The most relevant existing coverage sits inside cyber insurance policies, not property policies. Many cyber policies written since 2022 include a business interruption or system failure extension that is deliberately drafted without a physical damage requirement: the trigger is an outage, degradation, or failure of a computer system used in the business, whether caused by a malicious attack, an unintentional error, or in some wordings a third-party service failure. This is the correct policy family to check first, because it was built for exactly the kind of intangible, software-driven interruption an AI agent failure represents.
The practical caveat is scope. Not every cyber system failure extension covers every cause of AI agent downtime. Some are drafted narrowly around security incidents, meaning they respond to a ransomware attack or a hack but not to an ordinary model provider outage or a bad internal deployment. Others are drafted more broadly around any system failure regardless of cause. The only way to know which version you have is to read the trigger language in your own policy or ask your broker directly whether an AI agent outage with no malicious cause would be covered, not just a cyberattack.
Contingent and dependent business interruption
Contingent business interruption traditionally covers your loss of income following physical damage at a named supplier's premises: your key manufacturer burns down, and you cannot get the parts you need. Applied to an AI agent, the analogous scenario is your model provider's infrastructure going down. Standard contingent business interruption wording will not respond to this, for the same physical damage reason that defeats a direct claim, because a cloud provider's API outage does not involve physical damage at the provider's premises in any way the policy recognises.
A smaller number of cyber policies have begun writing dependent business interruption extensions that specifically name cloud service providers, SaaS platforms, or API dependencies as covered third parties, with a trigger based on the third party's system availability rather than physical damage. If your AI agent has a single point of failure at one model provider, this is the specific extension to ask about, and the specific question to ask is whether that model provider needs to be named in the policy schedule for the extension to apply, since many versions only cover pre-listed dependencies.
Parametric AI performance products
A distinct category of product avoids the physical damage question by design. Munich Re's aiSure product settles against an agreed performance specification: cover responds when the AI system's measured performance or availability falls below a specification agreed at the time the policy is placed, without requiring the policyholder to prove a causal chain from a covered peril to physical damage to loss.[3] This structure is well suited to downtime specifically, because uptime and performance against a threshold are the exact metrics a performance specification is built to measure. It is a specialist product, placed through brokers with the relevant delegated authority, and coverage capacity and eligibility depend on the specification quality an operator can provide at placement, which is a heavier lift for a small operator than a standard SME insurance purchase.
How to size your actual exposure before you shop for cover
Before approaching a broker about AI agent downtime coverage, calculate your own exposure rather than relying on an industry-wide figure, because the range across business types is too wide for a single number to be honest. Three questions produce a usable estimate.
First, what share of your revenue-generating activity currently runs through the AI agent without a fallback. A business where the agent handles order intake with no other channel faces a close-to-total revenue stop during an outage. A business where the agent supplements a staffed phone line faces a smaller, partial loss.
Second, what is your realistic outage duration. Model provider outages are typically measured in minutes to a few hours; a bad internal deployment you have to roll back manually can take longer to resolve, particularly outside business hours if nobody is monitoring the system continuously. Your realistic worst case, not your best case, is the number that matters for a coverage decision.
Third, what does your manual workaround actually cost, in overtime staffing, in lost customers who do not return, and in the reputational cost of a public-facing failure. This is often the largest and least visible component of the total loss, and it is precisely the extra expense element that a well-drafted business interruption extension is designed to pay alongside the direct lost income.
What to check in your existing policy this week
You do not need to buy a new product to take the first useful step. Ask your broker or insurer four specific questions about your current cyber and property policies. Does the business interruption or system failure extension in my cyber policy require a malicious cause, or does it respond to an ordinary technical failure with no attacker involved. Does the extension name my specific model provider or cloud dependency, or is it written broadly enough to cover any third-party system I rely on. What is the waiting period before the extension begins to pay, since many cyber business interruption extensions include a deductible period of eight to twelve hours before coverage activates, which matters if your realistic outage duration is shorter than that. And what is the maximum indemnity period, since a policy that pays for the first 30 days of interruption is of limited use if your realistic recovery timeline for a serious platform migration is longer.
If the answers reveal a gap, and for most SME operators running a customer-facing AI agent without a specific extension, they will, the practical next step is not necessarily a new standalone product. It is a conversation with your broker about adding or widening a non-physical business interruption extension on your next renewal, priced against the specific revenue exposure you calculated above rather than a generic AI risk premium.
Frequently asked questions
Does business interruption insurance cover AI agent downtime?
Usually not. Standard business interruption insurance is written on a physical damage trigger: it pays for lost income following direct physical loss or damage to insured property, such as a fire or a burst pipe. An AI agent going down because of a model provider outage, a bad deployment, or a software fault does not involve physical damage to anything you own, so most standard commercial property and business interruption policies will not respond, even if the outage stops your business from taking orders or serving customers for hours or days.
What does cover lost income from an AI agent outage?
The closest available coverage is a non-physical-damage business interruption extension, sometimes called a cyber business interruption or system failure extension, written into a cyber insurance policy rather than a property policy. These extensions pay lost income and extra expense following an outage of a computer system, including in some cases a third-party cloud or API dependency, without requiring physical damage. A small number of dedicated AI performance products, including Munich Re's aiSure, settle against a measured specification tied directly to system downtime or performance shortfall rather than a damage requirement.
What is a physical damage trigger and why does it exclude AI outages?
A physical damage trigger is the standard precondition in commercial property and business interruption policies: the insurer only pays for interrupted income if it follows direct physical loss or damage to covered property at a covered location, such as fire, flood, or storm damage to your premises or equipment. An AI agent failure, whether caused by a model provider outage, a corrupted deployment, or a configuration error, does not damage any physical object. Courts in several jurisdictions have confirmed that data loss and software failure alone do not meet a physical damage trigger, most notably in COVID-19 business interruption litigation, which tested this exact question at scale and largely found against policyholders relying on standard property triggers.
Does contingent business interruption cover a model provider outage?
Contingent business interruption traditionally covers loss of income caused by physical damage at a supplier's premises, for example a fire at your key manufacturer. It was not designed for a cloud API outage at your model provider, and standard wording will not respond to that scenario for the same physical damage reason that defeats a direct claim. Some cyber policies now write a specific dependent business interruption or system failure extension that names cloud and API providers as covered dependencies. If your AI agent depends on a single external model provider, confirming whether your cyber policy includes this extension, and whether that provider is a named or unnamed dependency, is a specific question to raise with your broker.
How much income can an AI agent outage actually cost a small business?
It depends entirely on how central the agent is to revenue generation. A business whose AI agent handles order-taking, appointment booking, or first-line sales qualification can lose the full run-rate of those functions for every hour the agent is down, plus the cost of manual workaround staffing and any customers who do not return. A business using an AI agent for a supporting function, such as internal drafting or back-office triage, faces a smaller and slower-building loss. There is no invented industry-wide average figure that applies honestly across business types; the right approach is to calculate your own hourly revenue dependency on the agent before assuming a number.
Related reading
Run the Coverage Audit
Before you talk to a broker, use the Coverage Audit tool to map your current policies against your AI agent exposure, including the business interruption gap this article describes. It takes ten minutes and produces the document your broker needs to review your position.
Start the Coverage AuditFootnotes
- ElevenLabs AI voice agent insurance placement, February 2026, structured under the AIUC-1 framework. Widely reported as the first company to insure its AI agents on a basis explicitly compared to employee coverage.
- Financial Conduct Authority v. Arch Insurance (UK) Ltd and others, UK Supreme Court, [2021] UKSC 1. The test case on non-damage business interruption extensions and COVID-19 government-ordered closures, establishing that only policies with specific non-physical-damage wording respond to non-physical interruption events.
- Munich Re, aiSure product framework. AI performance insurance settling against predefined specification thresholds rather than on a physical damage or causation-based trigger.
- For the wider policy-by-policy analysis of AI exclusions in cyber and errors and omissions cover, see the dedicated article on this site and on agentliability.eu.
- For carrier-level detail on parametric and non-parametric AI coverage products available in the European market, see the European AI agent insurance market tracker on agentinsured.eu.