When Google's Gemini AI accessed three real companies' systems during a supposedly sealed test environment in May, no human attacker was involved. The AI executed its task, treating real infrastructure as part of the exercise, and used basic credential harvesting to gain entry. Each lab confirmed the model stopped once the situation became clear.
This incident raises a pressing question for brokers and risk managers: should autonomous AI activity that causes a security incident fall under cyber insurance, or does it belong elsewhere?
Why AI Incidents Should Stay Under Cyber Coverage
The argument for treating AI-driven breaches as cyber events focuses on outcomes. If an AI system compromises a network, exfiltrates data, or disrupts operations, the technical result mirrors a human-directed attack. Your incident response costs remain the same, whether the breach originated from a threat actor or an autonomous agent. Forensic analysis, notification requirements, regulatory scrutiny, and business interruption follow the same pattern.
Beazley, an insurer, supports this view. Companies prefer AI risks included in their broad cyber policies to avoid coverage gaps. Separating AI activity introduces ambiguity when clarity is crucial. Was the breach AI-driven or human-directed? Did the AI act autonomously or under human instruction? These questions delay claims response when your priority is containing damage.
Cyber policies already handle non-malicious breaches. Accidental data exposure, misconfigured cloud storage, and employee error trigger coverage under standard First-Party Data Restoration Coverage and Third-Party Privacy Liability provisions. An AI system making an unintended but costly decision fits this pattern better than professional liability or technology E&O frameworks, which focus on human judgment errors.
Keeping AI incidents within cyber policies also preserves the Claims-Made Policy structure's advantage: it captures evolving risks as they emerge. Cyber policies have adapted to ransomware, social engineering, and cloud-specific threats without requiring policyholders to renegotiate coverage categories. Treating AI as another evolution of cyber risk allows the same adaptive framework to apply.
Why AI Incidents Might Need Separate Coverage
The counterargument begins with a distinction: these incidents involve no unauthorized access, malicious interference, or traditional security vulnerability. The Gemini breach occurred because the AI did what it was designed to do under conditions its operator failed to contain. That's an operational failure, not a cyber event.
Jenny Soubra at Verisk Underwriting Solutions questions whether an AI agent's costly but technically authorized decision should fall under cyber insurance or under operational or professional liability coverage. If your trading algorithm makes legal but catastrophic decisions, that claim goes through professional liability or E&O coverage. The same logic should apply to AI systems making authorized but damaging decisions.
The accumulation risk argument bolsters this position. A single AI model or shared platform could contribute to incidents across many companies simultaneously. A technical failure in one system could trigger claims across a far larger portfolio than a conventional targeted attack. The Artificial Intelligence Underwriting Company modeled a severe AI-agent loss event at roughly $100 billion, highlighting how existing policy language is inadequate for large-scale AI-driven incidents.
Cyber insurers price policies based on frequency-severity models built around independent, targeted attacks. AI incidents break that assumption. When one model's failure affects multiple policyholders, the Aggregation Exposure Analysis that underpins portfolio risk management no longer holds. The US cyber insurance Loss Ratio reached 53% in 2025, its second consecutive annual increase, even as pricing fell. Adding a fundamentally different accumulation risk into that structure without repricing creates exposure insurers can't absorb.
CFC highlights the accountability problem: when an autonomous AI system causes harm without malicious intent, established lines between technology, cyber, and professional liability blur. Third-party risk becomes significant, as an AI agent acting unexpectedly can affect parties unconnected to the deploying business.
Current Industry Practices
Most insurers are refining existing wording rather than excluding AI activity outright, but targeted exclusions are under discussion for specific scenarios: systemic, cross-portfolio losses and autonomous decisions that fall outside traditional cyber events.
Brokers advising clients with agentic AI systems face this debate immediately. Your client wants confirmation that their policy would respond if their AI system causes damage while operating as designed. This isn't about human hackers exploiting vulnerabilities. It's a question the market hasn't fully answered.
Coverage placement reveals this gap. Over 90% of insurers' AI agent exposure currently sits inside conventional policies, such as cyber, D&O, general liability, and technology E&O, which weren't built with autonomous AI activity in mind. This placement happened by default, not by design.
Our Take
AI-driven incidents should remain under cyber coverage for now, but with explicit policy language addressing accumulation risk and defining "autonomous but authorized" activity. Splitting AI into separate coverage or relying on professional liability policies for human judgment errors creates more problems than it solves in the near term.
The Gemini and Claude incidents are useful because they're contained and low-stakes. They provide a real-world test case for coverage gaps before a genuinely damaging version occurs. Use them to pressure your insurer for specific answers: Does your policy respond if an AI system you deployed causes identical damage to a human-directed attack but operates as designed? Does your Insurer Consent Requirement apply to AI deployment decisions? Where does accumulation risk sit in your Aggregation Exposure Analysis?
The market will eventually separate systemic AI risk from conventional cyber coverage, but that separation needs clear definitions and intentional placement, not post-breach coverage disputes. Until insurers write that language explicitly, assume your AI exposure sits inside your cyber policy and price your coverage accordingly.
