Many businesses now leverage AI as part of their day-to-day operations: drafting emails, summarizing documents, writing code.
But AI systems can also be designed to plan or select steps toward a goal and use tools to take actions. These capabilities exist on a spectrum: A system might simply recommend a next step, or it might execute transactions, alter systems, or carry out processes with limited human approval or supervision.
This additional autonomy forges novel opportunities for enterprises, but it also shifts the risk conversation into new territory. Brokers must now understand whether a client uses AI in the first place, but if so then what the system can access, what decisions it can make, and what happens if it gets something wrong.
What is agentic AI?
“Agentic AI” describes systems that can plan or select steps toward a goal and use tools or software to take actions.
Rather than being a single, clearly defined category, agentic AI exists on a spectrum. Some systems may recommend next steps without taking action, while others can execute transactions, alter systems, or carry out processes with limited human approval or supervision.
An “agent” in this regard may gather information, determine what to do next, use tools or software, and work through a series of tasks with varying degrees of human involvement. For example, an AI customer service agent might review a customer’s history, determine how to handle their problem, update the system, and arrange a follow-up.
The key consideration is therefore the degree of autonomy and authority a system has. The more an AI system can select actions and execute them, the more important its controls, permissions, and potential consequences become.
How are businesses using agentic AI?
Customer service: Triaging inquiries, accessing records, resolving straightforward issues, escalating exceptions
Operations: Monitoring workflows, identifying bottlenecks, allocating tasks
Research: Exploring sources, comparing information, producing recommendations
Software development: Writing and testing code, potentially initiating deployment processes
Finance: Reconciling transactions, flagging anomalies, preparing reports
The attraction is clear: Businesses can automate repetitive work and accelerate processes. But with greater autonomy comes the risk that an error travels further before being noticed.
Autonomous AI systems are introducing novel risks
Agentic AI doesn’t necessarily produce entirely new categories of risk. Rather, it may change the scale, speed, and pathway of existing risks.
Accountability
If an agent makes a decision that causes financial or operational harm, who is responsible? The business, its software provider, the employee overseeing the system, or another party altogether?
Inaccurate or unintended actions
An agent can act on incorrect information, misunderstand an objective, or take an action that was technically possible but commercially inappropriate.
Cyber, privacy, and IP risk
An AI system with access to internal applications, customer data, or financial systems produces another route through which a compromise could cause harm. Agents may also access sensitive information or external sources in ways that foster concerns surrounding privacy, IP, or confidentiality.
Professional and financial liability
When AI supports professional advice or business-critical decisions, a mistake could result in a claim. The risk may then be amplified further: While human error may affect one transaction, an automated agent could repeat the same mistake across thousands.
What does autonomous AI change for insurance?
When AI can gather information, make decisions, and act with only limited human supervision, who is liable? Who is accountable? That’s what CFC explored in our recent analysis.
Traditional risk management may no longer be enough
Traditional controls still matter, but businesses may need to adapt them for autonomous systems.
A practical AI risk management framework should consider:
governance: Who owns the AI system and approves its use?
access controls: What data and applications can it access?
human oversight: Which decisions require approval?
testing: Has the system been checked for unexpected behavior?
monitoring: Can unusual activity be detected quickly?
accountability: Is there a clear record of who designed and oversees it?
contingency planning: Can the business stop the agent and continue operating?
The more autonomous the system, the more important these controls become.
Insurance considerations for agentic AI
The insurance response depends on what the AI does, what can go wrong, and which policy comes into play.
Cyber insurance
Cyber coverage like CFC’s may be relevant where an agent is compromised, enables unauthorized access, exposes data, or contributes to system interruption; in other words, where the loss is fundamentally a cyber loss and AI is a contributing factor.
But where an agent takes an action that is erroneous but authorized, the applicable coverage may depend on the nature of the resulting loss and the policy wording. Depending on the circumstances, this could involve professional liability, technology E&O, management liability, or another applicable coverage.
Professional and technology liability
When AI is used to deliver technology services or professional work, an inaccurate output or automated action could lead to an allegation of negligence. Professional liability and technology E&O may therefore be relevant.
Management liability
If AI is involved in material business decisions, questions of oversight and accountability may extend to directors and officers. The more significant the decisions delegated, the more important governance becomes.
Intellectual property
AI-generated or AI-assisted outputs can raise questions of ownership, infringement, and third-party material. Brokers should consider whether existing policies address the client’s particular exposure.
Affirmative AI coverage
Clients shouldn’t assume that a standard policy automatically covers every AI-related loss. At CFC we advocate for affirmative AI coverage and specialist underwriting as AI use becomes ever more widespread. The key is always to assess the actual exposure rather than treat AI as a standalone risk category.
How can brokers help clients prepare?
Brokers can turn a broad conversation about AI into practical questions of risk:
Where is the client using AI?
Is it generating content, making recommendations, taking actions?
What systems and data can it access?
Can it make decisions without human approval?
What financial authority or operational control does it have?
Who monitors it?
How are errors detected and corrected?
What happens if the system is unavailable?
Which third parties supply the AI?
Does the current insurance program clearly address the relevant AI exposures?
These questions can establish whether AI is simply a tool of productivity or has become part of a business-critical process.
An agent that drafts internal notes presents a very different risk profile from one that can approve payments, change production systems, or provide advice to customers. Brokers should therefore also encourage clients to document their AI use, including systems, permissions, data access, decision authority, and human oversight.
What will agentic AI mean for insurance?
Agentic AI is set to render insurance risk far more dynamic. As businesses delegate more tasks to autonomous systems, the consequences of failure will surely become faster, broader, and harder to attribute.
Autonomous AI is shifting the conversations about accountability, third-party risk, professional responsibility. This points toward a more specific approach to AI underwriting moving forward, focused on how technology is actually being used rather than simply whether a business uses AI.
For brokers, this represents an opportunity. After all, clients may not recognize that an AI system has become an operational dependency, or that its permissions produce a material exposure. So asking the right questions can help uncover those risks before they turn into claims.
A practical starting point for brokers
Agentic AI is so much more than an “advanced version” of generative AI. Its defining feature is autonomy: the ability to pursue tasks, make decisions, and take actions with limited human intervention.
In this way, agentic AI is spawning novel business opportunities, but also changing the risk profile. Errors can be amplified, accountability can become unclear, and systems with broad access can produce cyber, privacy, professional, and financial exposures.
None of this is to say that brokers need to become AI engineers overnight. Instead they must just understand what their clients’ systems can do, what authority they have, and what happens when they fail. And if you would like to discuss AI risk or how insurance can respond to emerging technology exposures, get in touch with CFC today to explore the options for your clients.
