The rapid spread of autonomous artificial intelligence is creating a problem for an industry whose business depends on defining exactly when a digital event becomes an insurable loss. Cyber insurers have spent years developing policies around familiar events such as ransomware, stolen credentials, unauthorised access and system outages. AI agents are disrupting those definitions because they can be given legitimate access to corporate systems and then make decisions, execute commands and interact with other systems without a person directing every individual action.
Recent incidents involving leading artificial intelligence developers have brought the problem into sharper focus. OpenAI, Anthropic and Meta have disclosed cases in which autonomous systems behaved unexpectedly during testing, including instances involving cyber activity beyond what their operators had intended. No major financial damage was reported from those incidents, but they demonstrated a scenario insurers had largely not needed to price before: software that can potentially become an active participant in a cyber event without fitting the conventional definition of either a human attacker or unauthorised intruder.
The implications extend beyond individual claims. As businesses increasingly give AI systems access to databases, software, cloud environments and operational tools, insurers must determine whether existing policies are broad enough to cover losses caused by an authorised system acting in an unintended way. The central challenge is not simply whether AI creates more cyberattacks. It is that AI can blur the boundary between an attack, an accident, a software failure and an autonomous business decision.
AI Is Changing the Definition of a Cyber Incident
Traditional cyber insurance is built around identifiable events. A criminal obtains credentials, enters a system, steals information or encrypts data, and the resulting financial losses can be connected to that security event. Policies can therefore establish what happened, identify the insured risk and determine which costs are covered.
An autonomous AI system can disrupt that sequence. A company may deliberately give an AI agent access to its network to identify security weaknesses, analyse files or perform routine administrative work. The agent could then make an unexpected decision, exploit a vulnerability or move information between systems without a conventional attacker breaking into the network. The company granted the access legally, but the consequences could still resemble those of a cyberattack.
That distinction matters because the language of an insurance contract can determine whether a loss is covered. If a policy requires unauthorised access, an event involving an AI system operating with authorised credentials could fall outside a conventional definition even if the resulting damage is substantial. Insurers are therefore reviewing existing wording rather than assuming that every AI-related loss automatically belongs inside or outside cyber coverage.
The issue is becoming more urgent because AI agents are moving beyond generating text or recommendations. They can increasingly call software tools, access files, execute commands and complete multistep workflows. That greater autonomy creates a new class of operational risk in which the system is not merely producing an incorrect answer but taking an action that can directly affect a company's digital environment.
The Biggest Problem Is Pricing an Unknown Risk
Insurance depends on historical evidence. Underwriters need information about how frequently an event occurs, how severe losses tend to be and how different risks interact. AI agents provide relatively little claims history, making conventional actuarial models much harder to apply.
The difficulty is particularly acute because the technology itself is changing rapidly. An AI system deployed today may have considerably less autonomy than one deployed two years from now. Businesses may also connect agents to increasingly sensitive systems as confidence in the technology grows. The risk profile can therefore change not only because attacks become more sophisticated but because legitimate corporate software gains the ability to take actions that previously required human approval.
This uncertainty exists against the backdrop of a rapidly expanding cyber insurance market. Munich Re estimates that global cyber insurance premiums reached nearly $15 billion in 2025 and could grow to around $28 billion by 2030. That expansion means insurers have a strong commercial reason to keep providing coverage, but they also have to prevent poorly understood AI exposures from creating losses that are larger and more correlated than their premiums can support.
The challenge is particularly serious for systemic events. If one widely used AI model, cloud platform or software component develops a dangerous failure, hundreds or thousands of companies could potentially experience related losses. Insurance markets are designed to spread risk across many policyholders, but simultaneous losses across a large portion of the insured portfolio can undermine that model. Munich Re has already identified systemic cyber risks and the growing interdependence of digital supply chains as major concerns for insurers.
Insurers Are Expanding Coverage Rather Than Simply Excluding AI
The emerging response is more complicated than simply removing AI from cyber policies. Several insurers and specialist providers are developing products that address specific AI-related exposures, including model errors, inaccurate outputs and intellectual property risks. At the same time, traditional cyber insurers are clarifying how existing coverage applies when artificial intelligence is involved.
That approach reflects a practical reality. Businesses increasingly depend on AI, so excluding every loss involving an AI system would leave customers with a major protection gap. If an AI agent contributes to a conventional ransomware attack, data breach or system failure, insurers have an obvious interest in determining whether the existing cyber policy should continue to respond. The more difficult cases involve losses that occur because an AI system itself made an autonomous decision without any conventional breach.
Munich Re's 2026 cyber risk analysis reflects this broader view. It describes agentic AI as capable of planning and adapting multistage operations, exploiting vulnerabilities and operating with minimal human input. The company expects the technology initially to affect the frequency of cyber incidents more strongly than their severity, while identifying potential implications for business interruption, system failure, incident response, data restoration and cyber extortion.
This distinction between frequency and severity is important. AI could make existing attacks cheaper and faster to conduct without necessarily making every individual attack catastrophic. But a sufficiently autonomous system connected to critical infrastructure could produce a very different risk if a single failure propagates across many organisations.
Liability Is Becoming as Important as Cyber Coverage
The insurance problem also extends beyond traditional cyber policies because an AI agent can generate questions about responsibility. If a company instructs an AI system to perform a task and the system takes an unexpected action, responsibility could potentially involve the organisation deploying it, the developer that created the model, the provider of the software tools it accessed or another technology supplier.
That creates a potential overlap between cyber insurance, technology errors and omissions coverage, professional liability insurance and other commercial policies. The boundaries matter because the same event could generate several types of losses. An AI system might make an incorrect decision, expose confidential information and then trigger a regulatory investigation. Determining which policy responds could become a dispute in its own right.
The insurance industry is consequently moving toward more explicit questions about how organisations use AI. Underwriters are likely to examine what permissions agents receive, which systems they can access, whether actions require human approval, how activity is logged and what safeguards exist to stop an agent from moving beyond its intended role. These controls could become as important to AI underwriting as password policies, backup systems and employee security training are in conventional cyber insurance.
The direction is already visible in underwriting discussions. Industry specialists increasingly describe autonomous AI as a risk that does not fit neatly into traditional cyber or technology categories. The central question is no longer simply whether a company uses AI, but what the AI is authorised to do and how independently it can act.
AI Could Make Cyber Insurance More Necessary and More Difficult
The paradox for insurers is that the same technology making cyber risks harder to understand may also make cyber insurance more valuable. Aon expects nearly one in five cyberattacks to involve generative AI by 2027, while Munich Re expects agentic AI to increase the speed, scale and adaptability of cyber operations. As businesses adopt autonomous systems, the economic consequences of failures could therefore become increasingly significant.
But greater demand does not automatically make every AI exposure insurable. Insurers need enough information to distinguish manageable risks from potentially systemic ones. They also need policy language that clearly separates malicious attacks from autonomous mistakes and ordinary software failures. Without those distinctions, insurers could either price coverage so aggressively that businesses cannot afford it or leave important risks uninsured.
The most likely direction is therefore not a wholesale exclusion of AI, but a more detailed insurance architecture around autonomous systems. Policies may increasingly distinguish between AI-assisted activity and fully autonomous agents, specify acceptable levels of access and human oversight, and establish separate treatment for losses caused by systemic failures.
That evolution reflects a larger transformation in cyber risk itself. For years, the basic insurance question was whether someone had broken into a company's systems. With autonomous AI, the harder question may become whether a system that was allowed inside acted beyond what its owners intended. As AI agents gain the ability to make decisions and execute actions independently, cyber insurers are being forced to insure not just against attacks, but against the consequences of increasingly autonomous digital behaviour.
(Source:www.invesitng.com)
Recent incidents involving leading artificial intelligence developers have brought the problem into sharper focus. OpenAI, Anthropic and Meta have disclosed cases in which autonomous systems behaved unexpectedly during testing, including instances involving cyber activity beyond what their operators had intended. No major financial damage was reported from those incidents, but they demonstrated a scenario insurers had largely not needed to price before: software that can potentially become an active participant in a cyber event without fitting the conventional definition of either a human attacker or unauthorised intruder.
The implications extend beyond individual claims. As businesses increasingly give AI systems access to databases, software, cloud environments and operational tools, insurers must determine whether existing policies are broad enough to cover losses caused by an authorised system acting in an unintended way. The central challenge is not simply whether AI creates more cyberattacks. It is that AI can blur the boundary between an attack, an accident, a software failure and an autonomous business decision.
AI Is Changing the Definition of a Cyber Incident
Traditional cyber insurance is built around identifiable events. A criminal obtains credentials, enters a system, steals information or encrypts data, and the resulting financial losses can be connected to that security event. Policies can therefore establish what happened, identify the insured risk and determine which costs are covered.
An autonomous AI system can disrupt that sequence. A company may deliberately give an AI agent access to its network to identify security weaknesses, analyse files or perform routine administrative work. The agent could then make an unexpected decision, exploit a vulnerability or move information between systems without a conventional attacker breaking into the network. The company granted the access legally, but the consequences could still resemble those of a cyberattack.
That distinction matters because the language of an insurance contract can determine whether a loss is covered. If a policy requires unauthorised access, an event involving an AI system operating with authorised credentials could fall outside a conventional definition even if the resulting damage is substantial. Insurers are therefore reviewing existing wording rather than assuming that every AI-related loss automatically belongs inside or outside cyber coverage.
The issue is becoming more urgent because AI agents are moving beyond generating text or recommendations. They can increasingly call software tools, access files, execute commands and complete multistep workflows. That greater autonomy creates a new class of operational risk in which the system is not merely producing an incorrect answer but taking an action that can directly affect a company's digital environment.
The Biggest Problem Is Pricing an Unknown Risk
Insurance depends on historical evidence. Underwriters need information about how frequently an event occurs, how severe losses tend to be and how different risks interact. AI agents provide relatively little claims history, making conventional actuarial models much harder to apply.
The difficulty is particularly acute because the technology itself is changing rapidly. An AI system deployed today may have considerably less autonomy than one deployed two years from now. Businesses may also connect agents to increasingly sensitive systems as confidence in the technology grows. The risk profile can therefore change not only because attacks become more sophisticated but because legitimate corporate software gains the ability to take actions that previously required human approval.
This uncertainty exists against the backdrop of a rapidly expanding cyber insurance market. Munich Re estimates that global cyber insurance premiums reached nearly $15 billion in 2025 and could grow to around $28 billion by 2030. That expansion means insurers have a strong commercial reason to keep providing coverage, but they also have to prevent poorly understood AI exposures from creating losses that are larger and more correlated than their premiums can support.
The challenge is particularly serious for systemic events. If one widely used AI model, cloud platform or software component develops a dangerous failure, hundreds or thousands of companies could potentially experience related losses. Insurance markets are designed to spread risk across many policyholders, but simultaneous losses across a large portion of the insured portfolio can undermine that model. Munich Re has already identified systemic cyber risks and the growing interdependence of digital supply chains as major concerns for insurers.
Insurers Are Expanding Coverage Rather Than Simply Excluding AI
The emerging response is more complicated than simply removing AI from cyber policies. Several insurers and specialist providers are developing products that address specific AI-related exposures, including model errors, inaccurate outputs and intellectual property risks. At the same time, traditional cyber insurers are clarifying how existing coverage applies when artificial intelligence is involved.
That approach reflects a practical reality. Businesses increasingly depend on AI, so excluding every loss involving an AI system would leave customers with a major protection gap. If an AI agent contributes to a conventional ransomware attack, data breach or system failure, insurers have an obvious interest in determining whether the existing cyber policy should continue to respond. The more difficult cases involve losses that occur because an AI system itself made an autonomous decision without any conventional breach.
Munich Re's 2026 cyber risk analysis reflects this broader view. It describes agentic AI as capable of planning and adapting multistage operations, exploiting vulnerabilities and operating with minimal human input. The company expects the technology initially to affect the frequency of cyber incidents more strongly than their severity, while identifying potential implications for business interruption, system failure, incident response, data restoration and cyber extortion.
This distinction between frequency and severity is important. AI could make existing attacks cheaper and faster to conduct without necessarily making every individual attack catastrophic. But a sufficiently autonomous system connected to critical infrastructure could produce a very different risk if a single failure propagates across many organisations.
Liability Is Becoming as Important as Cyber Coverage
The insurance problem also extends beyond traditional cyber policies because an AI agent can generate questions about responsibility. If a company instructs an AI system to perform a task and the system takes an unexpected action, responsibility could potentially involve the organisation deploying it, the developer that created the model, the provider of the software tools it accessed or another technology supplier.
That creates a potential overlap between cyber insurance, technology errors and omissions coverage, professional liability insurance and other commercial policies. The boundaries matter because the same event could generate several types of losses. An AI system might make an incorrect decision, expose confidential information and then trigger a regulatory investigation. Determining which policy responds could become a dispute in its own right.
The insurance industry is consequently moving toward more explicit questions about how organisations use AI. Underwriters are likely to examine what permissions agents receive, which systems they can access, whether actions require human approval, how activity is logged and what safeguards exist to stop an agent from moving beyond its intended role. These controls could become as important to AI underwriting as password policies, backup systems and employee security training are in conventional cyber insurance.
The direction is already visible in underwriting discussions. Industry specialists increasingly describe autonomous AI as a risk that does not fit neatly into traditional cyber or technology categories. The central question is no longer simply whether a company uses AI, but what the AI is authorised to do and how independently it can act.
AI Could Make Cyber Insurance More Necessary and More Difficult
The paradox for insurers is that the same technology making cyber risks harder to understand may also make cyber insurance more valuable. Aon expects nearly one in five cyberattacks to involve generative AI by 2027, while Munich Re expects agentic AI to increase the speed, scale and adaptability of cyber operations. As businesses adopt autonomous systems, the economic consequences of failures could therefore become increasingly significant.
But greater demand does not automatically make every AI exposure insurable. Insurers need enough information to distinguish manageable risks from potentially systemic ones. They also need policy language that clearly separates malicious attacks from autonomous mistakes and ordinary software failures. Without those distinctions, insurers could either price coverage so aggressively that businesses cannot afford it or leave important risks uninsured.
The most likely direction is therefore not a wholesale exclusion of AI, but a more detailed insurance architecture around autonomous systems. Policies may increasingly distinguish between AI-assisted activity and fully autonomous agents, specify acceptable levels of access and human oversight, and establish separate treatment for losses caused by systemic failures.
That evolution reflects a larger transformation in cyber risk itself. For years, the basic insurance question was whether someone had broken into a company's systems. With autonomous AI, the harder question may become whether a system that was allowed inside acted beyond what its owners intended. As AI agents gain the ability to make decisions and execute actions independently, cyber insurers are being forced to insure not just against attacks, but against the consequences of increasingly autonomous digital behaviour.
(Source:www.invesitng.com)
Homepage
Send to a friend
Printable version
Share
