Photo By: Matheus Farias
Companies are racing to adopt artificial intelligence. But before they ask what AI can do, they may need to answer a simpler question: What AI do we actually have?
Artificial intelligence is becoming part of everyday business. Companies are using AI to write emails, analyze data, create marketing materials, answer customer questions and help employees complete tasks.
But while businesses are paying close attention to the AI they officially approve, another type of AI may be quietly spreading inside their organizations.
Employees may be using personal AI accounts for work. Developers may be experimenting with AI coding tools. Teams may sign up for AI services without going through their IT departments. Some employees may even create AI agents that can perform tasks automatically.
This creates a growing problem for businesses.
They may not know how much AI is actually operating inside their organization.
For years, companies have worked to keep track of their computers, software, cloud services and user accounts. Security teams know that they need to understand what is connected to their networks and who has access to important information.
AI adds another layer to that challenge.
Companies now need to ask a different question:
What AI systems are acting on our behalf?
The Rise of the AI Agent
The term “AI agent” is becoming increasingly common as companies look for ways to automate work.
A chatbot generally responds when a person asks it a question. An AI agent can potentially do much more.
Depending on how it is designed, an agent can be given a goal and then take several steps to achieve it. It might search for information, work with files, use other software, write code or complete tasks without a person telling it what to do at every step.
That can be extremely useful for businesses.
It can also create new risks.
An agent with access to sensitive company information could potentially expose that information. An agent with access to business systems could make changes. An agent with access to email could send messages.
The more an AI system can do, the more important it becomes for a company to understand exactly what that system has access to.
What Makes an AI “Rogue”?
The idea of a rogue AI agent can sound dramatic.
It does not necessarily mean an AI system has become malicious or developed its own intentions.
A rogue agent could simply be an AI system operating without proper oversight.
For example, an employee might create an AI agent to automate a repetitive task. The project may start as a harmless experiment. The employee could then move to another role or leave the company, while the agent continues to operate.
Another employee might connect an AI tool to company files without realizing how much information the tool can access.
An organization could also approve an AI system for one purpose and later discover that it has access to systems or information it does not actually need.
In each case, the problem is the same.
The company does not have a clear picture of what the AI is doing.
The Problem With “Shadow AI”
Businesses are already familiar with the idea of “shadow IT.” This happens when employees use technology that has not been approved by the organization.
AI makes shadow IT much easier.
An employee can sign up for an AI service in minutes. They may not need help from IT or a software installation. They can simply start using it.
That convenience is one of AI’s biggest strengths.
It is also one of its biggest challenges for security teams.
An employee might upload a document to an AI tool because they want help summarizing it. They might connect an AI assistant to their calendar or email to save time. A developer might give an AI coding agent access to a project.
Each decision may seem small.
But when hundreds or thousands of employees make similar decisions, a company can quickly end up with an AI environment that is difficult to understand.
Companies Need an AI Inventory
The first step in dealing with this problem is visibility.
Companies need to know what AI systems are being used, who controls them and what they can access.
That could mean creating an AI inventory that includes both officially approved systems and AI tools being used by employees.
The inventory should answer basic questions:
What AI systems are being used?
Who owns each system?
What information can it access?
What company applications can it connect to?
Does it have access to passwords, accounts or other credentials?
Can it take actions without human approval?
Who monitors it?
When was it last reviewed?
Can it be turned off quickly if something goes wrong?
These questions may eventually become a normal part of corporate security.
Companies already track computers and software because they understand that unknown technology can create security problems.
AI should be treated in much the same way.
The Importance of Permissions
One of the biggest issues with AI agents may not be the AI itself. It may be what the AI is allowed to access.
An AI tool that summarizes publicly available information presents a very different risk from an AI agent that can access customer records, company databases and internal documents.
The more permissions an agent has, the more carefully it needs to be monitored.
This is where traditional security principles still matter.
AI agents should generally have only the access they need to perform their jobs. Companies should be able to monitor their activity, review their permissions and shut them down when necessary.
In other words, companies should not simply ask whether an AI system is smart.
They should ask:
What can it reach?
The Conversation Is Moving Beyond AI Adoption
These questions are becoming part of a larger conversation about how organizations manage AI.
On September 16, NewRocket and ServiceNow will bring this issue into focus during a breakfast discussion at the Calgary Petroleum Club called “On the Frontier of AI: An Opportunity.”
One of the central topics will be exactly the issue facing many organizations today: knowing what AI they have, including rogue agents. The discussion will also cover governance and compliance, measuring the value of AI, deciding where to invest next and how to put those investments into practice with confidence.
The event will look at the entire life of an AI asset, including how organizations can track and inventory tools such as Copilot, Claude, Gemini and internally built agents. It will also explore questions around ownership, accountability, risk and the lifecycle of AI systems.
The session will include real-world examples, followed by a question-and-answer session and networking with other business leaders. The breakfast begins at 7:30 a.m., with the discussion starting at 8:00 a.m. at the Calgary Petroleum Club.
The timing of that conversation reflects a broader change in how businesses are thinking about AI.
The question is no longer simply, “How can we use AI?”
It is becoming, “How do we know what AI we have, what it can do and whether we can control it?”
Knowing What You Have
The AI industry is moving quickly. New models, tools and agents are appearing all the time.
Businesses will continue to adopt them because they can save time, reduce costs and help employees work more efficiently.
But adoption without visibility creates a problem.
Before companies can properly manage their AI, they need to know where it is.
They need to know which systems are approved, which ones employees are using independently and which agents have the ability to take action.
The goal is not to stop employees from using AI.
The goal is to make sure the company understands what is happening.
The biggest AI risk inside an organization may not be the system that executives approved and security teams monitor every day.
It may be the AI nobody remembers approving.
Or worse, the AI nobody knows exists.