Your Most Productive Employees Might Be Your Biggest AI Security Risk

The short version

A new report found that a tiny fraction of employees—the top 5% of AI users—are creating huge security holes for their companies. These enthusiastic “super-adopters” use AI tools way more than anyone else, often on personal accounts, creating a “shadow AI” problem. This is a quiet reminder that what feels like harmless productivity can easily become a major data leak.

What is ‘shadow AI’ and why should I care?

You’ve probably heard of “shadow IT.” That’s when employees use software, apps, or services without the company’s approval or knowledge. It’s been happening for years—think using your personal Dropbox to share work files because it was easier than the clunky corporate server. “Shadow AI” is the 2024 version of that, but with much higher stakes.

Shadow AI refers to the use of artificial intelligence tools and platforms outside of a company’s official, sanctioned systems. This happens when an employee uses their personal ChatGPT account to summarize meeting notes, asks Claude to draft a sensitive email, or uses a random online AI tool to analyze sales data. On the surface, it’s just an employee trying to be more efficient. The problem is where that data goes. When you paste text into a public AI model, you’re essentially sending your company’s information to a third-party server. Your IT and security teams have zero visibility into what was shared, who saw it, or how it’s being used to train future models. It’s a massive governance and security blind spot.

This isn’t a theoretical problem. People do this every single day. The impulse is understandable; you have a task, and you know an AI tool can do it in seconds. The official, company-approved tool might be slower, less capable, or buried under a mountain of access requests. So you open a new tab, log into your personal account, and get the job done. The convenience is immediate, but the potential risk is invisible and long-lasting.

Who are these ‘super-adopters’?

This is where the new research gets really interesting. Akamai, a company that sees a huge amount of the world’s internet traffic, analyzed enterprise AI usage and uncovered a fascinating pattern. They identified a group they call “super-adopters,” which makes up the top 5% of AI users within a company. These aren’t malicious actors; they’re your most enthusiastic, AI-forward colleagues. They are the ones who are genuinely excited about this technology and are using it to be as productive as possible.

The data shows these super-adopters interact with AI models at a rate 12 times higher than the bottom 50% of users. They are driving the adoption curve, but because they’re moving so fast, they are also driving The outsized shadow AI security risk. The report found that nearly half of all enterprise AI interactions happen through personal accounts, not corporate-managed ones. That means for every two queries your company’s employees make to an AI, one of them is happening completely off the books, in a system the company can’t see or protect.

Think about that for a second. Half of the conversations your colleagues are having with AI—which could include pasting in customer data, internal strategy documents, unreleased marketing copy, or proprietary source code—are happening in the wild. These super-adopters aren’t trying to cause trouble. They’re just trying to do their jobs better and faster. But in the process, they are creating a security risk that is disproportionately larger than their small group size would suggest.

What new risks does shadow AI create?

Beyond the obvious risk of data leakage, the proliferation of shadow AI introduces new and frankly creative ways for attackers to cause problems. The Akamai report highlights a couple of spooky-sounding attack vectors: “Vibe Hacking” and “CursorJacking.” These aren’t just buzzwords; they represent real threats that exploit the human-AI interaction.

“Vibe Hacking” is a sophisticated form of social engineering. An attacker could compromise an AI tool and subtly alter its personality or “vibe” to seem more friendly, more urgent, or more authoritative. The goal is to manipulate the user into trusting the AI more than they should, eventually tricking them into revealing sensitive information they would normally hold back. Imagine an AI that suddenly starts acting like your boss, using a similar tone and phrasing, and then asks for a “quick summary of the quarterly numbers” before the official release. It’s a subtle but powerful manipulation.

“CursorJacking” is even more devious. This attack involves malicious code on a webpage that tracks the user’s cursor movements. While this sounds minor, consider how you interact with text. You often highlight sentences with your cursor as you read them, or hover over a specific piece of data before you copy and paste it. An attacker could use this cursor data to reconstruct sensitive information you’re viewing or preparing to paste into an AI prompt, even if you never actually hit “send.” They are, in effect, looking over your shoulder digitally and stealing data before it even leaves your machine.

These threats compound the core problem: when employees use unvetted AI tools, they aren’t just risking the data they paste in. They are also exposing themselves and their company to novel attacks designed specifically for this new way of working. The security features and monitoring that come with official tools like Microsoft Copilot or Gemini Enterprise are there for a reason—to protect against exactly these kinds of threats.

What should you do about this?

This isn’t just a problem for your IT department to solve with a bigger firewall. The reason shadow AI is so prevalent is because it’s a people problem, driven by the desire for productivity and convenience. So, the solution has to involve people, too.

If you’re an employee, the first step is awareness. Before you paste anything into a public AI tool, stop and think: “Would I be comfortable if this text was posted on a public forum?” If the answer is no, don’t paste it. This includes seemingly harmless things like internal emails, customer support chats, or draft documents. That information belongs to your company, and putting it into your personal AI account is like leaving a printout on a coffee shop table. Always default to using the company-sanctioned AI tools. If you don’t know what they are, ask your manager or IT. If the official tools are frustrating to use, provide that feedback constructively instead of just working around them.

If you’re a manager or run a company, you can’t just ban everything. Prohibition rarely works. Instead, you need to make the secure path the easy path. Invest in good, enterprise-grade AI tools that your employees will actually want to use. Educate your team not just on the rules, but on the reasons for the rules. Explain the risks of data leakage and attacks like Vibe Hacking in plain language. When people understand the “why,” they are much more likely to be your partners in security rather than adversaries. The goal is to enable productivity, not stifle it—but to do so within a secure framework that protects the company’s most valuable asset: its data.

FAQ

What is shadow AI? Shadow AI is the use of any artificial intelligence tools, applications, or platforms for work purposes without the explicit approval and oversight of a company’s IT and security departments. It commonly involves using personal AI accounts (like a free ChatGPT account) for work-related tasks.

Is it always bad to use personal AI tools for work? Yes, if you are handling any information that is not public knowledge. Pasting internal company data, customer information, or intellectual property into a public AI model on a personal account creates a significant data leak and security risk, as the company has no control over that data.

Why do employees use shadow AI if it’s so risky? Employees typically turn to shadow AI for convenience and efficiency. Their personal tools might be more powerful, faster, or more familiar than the official corporate-sanctioned alternatives, leading them to prioritize productivity over security protocols they may not fully understand.

What’s the difference between a personal and enterprise AI account? Enterprise accounts (like Microsoft Copilot or Gemini Enterprise) are managed by your company. They include critical security features, data privacy controls that prevent your information from being used for model training, and activity logs for security oversight. Personal accounts offer none of these protections for your work data.

How can a company prevent shadow AI? A combination of strategies is most effective. Companies should provide employees with powerful, easy-to-use sanctioned AI tools, educate the workforce on the specific risks of using unapproved tools, and establish clear policies that balance security with the need for productivity.