The unauthorized use of software and applications has been a problem for MSPs and security teams for decades. This "Shadow IT" is a hidden risk because the software doesn’t go through the proper vetting processes and may not meet company security problems. As users find new tools and applications to help simplify and speed up processes, new risks are introduced, and it's not always straightforward to monitor what unapproved software is being used or how to make sure sensitive data stays protected.
Then AI came along. And everything got much worse.
Why Shadow AI amplifies an existing problem
Not having visibility over what applications users are working with means there can be unknown risk vectors that could potentially compromise security. As many of these applications are purpose-built, they often don't have much access to company data. Employees aren't likely to share sensitive financial forecasts with an app that reminds them to take breaks every hour or proprietary code with a tool that helps send marketing emails.
AI changed the game.
The conversational nature, instant response, and near limitless applications make it a Swiss Army Knife that is useful across disciplines, industries, and departments. And then there’s there the accessibility. You simply ask it what you want in plain English then tune your prompts based on the response until you get what you need. That becomes a risk when employees start sharing organizational data with an unsanctioned tool because AI works much better with specific datasets.
Most of the time, this isn’t malicious. Employees typically are just looking for better ways to complete their tasks, and if the business doesn’t provide the AI service or application to handle it, the determined find a way to do it themselves. The problem is that these unsanctioned tools haven’t been analyzed for privacy and security, so can become risks for both cybersecurity, privacy, and compliance. We talk about this more in Managing sensitive data in AI.
The problem gets serious once sensitive data is shared with these unsanctioned AI tools, especially when it’s not always easy to find out how they’re being used.
How widespread is Shadow AI really?
At first glance, Shadow AI might seem like something contained to a small group of users, especially with sensitive company or customer data.
But look across any study or search expert opinions on Shadow AI, and you’ll see the same sentiment: it’s a huge problem and outpacing any solutions to contain it. Experts in the field such as Trusted Tech, PagerDuty, Varonis, and Proofpoint suggest that between ½ and ⅔ of employees use unapproved AI tools for work, and that most of these employees do so while knowing there are significant risks for privacy and data security.
This isn’t limited to new or non-management team members either. In fact, the Trusted Tech survey claims that senior decision-makers are more than twice as likely to use unapproved AI tools as their employees. These are experienced users with greater privileges and access to a wider set of sensitive, proprietary, and confidential data, who still choose to use these tools regardless. The more senior, the more likely to use unsanctioned tools, with the C-suite being the worst… and with the most to lose.
So why the discrepancy? Why are employees, especially ones with more responsibility, knowingly using these tools despite being cognizant of the risks? And is there anything that can be done to stop it?
Why employees are using unapproved AI tools
There are a lot of reasons users work with AI tools. But why are they using unauthorized ones? While individual motives vary, research has shown some similarities for different groups of employees.
New or lower-level employees:
- Fear that they won’t be able to use the tools they like and are familiar with
- Believe that they understand AI better than the team responsible for putting the corporate policies in place
- Want to avoid judgment by their coworkers or managers about using a specific (or any) tool.
Management and C-suite:
- Find the approved tools are limiting and less efficient
- Worry about surveillance on approved business tools
- Are concerned that too much use could negatively affect perception of competence at their job.
Both groups don’t want to lose their productivity gains and have a familiarity with consumer models for personal use. So when the approved tools aren't suited for the job, they find one that is. But is it possible to keep the benefits of AI without all the risks unapproved tools pose?
With the right tools, management, and steps, it can be.
How to find Shadow AI
Because Shadow AI often happens due to not knowing about the proper procedure/requirements or deliberately choosing to use tools that aren't approved, it's not always simple to find it. Surveying teams about the AI tools they're using is a good start, but it's likely to miss anyone who is intentionally hiding tool use.
Instead, a modern approach is to use a technical solution for finding and monitoring Shadow AI.
Solutions for Shadow AI
If the team is more efficient with AI, removing the tools that helped make them that way will hurt. Doing nothing could hurt a whole lot more. Letting users continue to use unapproved tools is risky, opening up the organization to exposing data to outside (or internal) parties and breaking compliance, which can result in hefty fees or losing clients.
There isn’t one surefire solution to this problem yet. But there are some options to help mitigate it, and tools that can help you detect and manage Shadow AI.
Typically, there are three main approaches: Ban it, educate about it, or monitor it.
Banning Shadow AI
A good idea in theory. Bring down the ban hammer and unauthorized tools from ever becoming a problem. If it’s not allowed, it’s over, right?
For some employees, yes. They won’t be able to use their work credentials to sign up for or use tools without approval.
And that’s the problem.
While you’ll be able to stop some Shadow AI with bans, the survivors get more dangerous. Instead of using locked-down, monitored work accounts, some will switch to private accounts. Or worse.
Sure, there are plenty of tools and tricks to ban unapproved tools. You can blacklist or whitelist specific websites. You can restrict company devices to only connect to company networks. You can monitor employee devices and remotely deactivate applications.
But that doesn’t get to the root of the problem. And a determined user will still find the way around it if it’s crucial to performing their duties. They might use a personal device, screenshot the sensitive details, type them in manually, or as was the case for one Reddit report, send simple requests to an unmonitored personal network and have the home device initiate the request to AI and then retrieve the response from the work device.
Banning can be effective for a large portion of cases with the right solutions in place. But unless you have a completely secure, air-gapped state-of-the-art facility that forces extreme security policies, banning without an explanation can drive people who want to use a specific tool will find a way to use it. And when they do, it will probably be a much more problematic way than using their work computer.
Education and employee training
People don’t know what they don’t know. So help them know what they need to know, you know? Giving your team the information they didn’t know can at the very least raise awareness.
Above, we mentioned that many employees are using unsanctioned tools even though they know the risks. But some employees genuinely don’t know.
For instance, there may be staff who aren’t aware that LLMs retain the information. To them, it might simply be a shortcut to get information quickly. If they’ve never been taught how AI tools work, how they process information, and the negative ramifications of unsanctioned tools, they can’t really be blamed, right?
Plus, knowing “a risk exists” versus specific risks and consequences that could result are very different things. Someone can know destroying a dam will cause some water to leak, but not that it ends up flooding thousands of homes. Education can help people understand the true extent of the problem and what can happen.
Company policies and training can help bridge the knowledge gap, but it doesn’t ensure the policies will actually be followed. Regular refreshers help, and work best when used alongside coinciding policies to explain why the policies are in place and how they protect users and data.
Monitoring for unapproved tools
An alternative to outright banning is monitoring. Find out what employees are actually doing. Get the whole picture. If you notice something is a problem, then you can take another action like banning it or educating the person using it.
You can learn a lot in the process too. Maybe you as a company invested in ChatGPT but most of your workers choose Claude instead. Maybe you discover a legitimately useful tool with a specific purpose that the wider team can adopt. Or maybe you find out the problem is much bigger (or not as big) as you thought.
Monitoring gives you visibility and the context to make decisions on what to do next.
Combating Shadow AI through monitoring can also help change the paradigm of AI usage in the workplace. For some users, simply knowing that their usage is visible will be enough to make them think twice about which tools they’re using. Even when there aren’t consequences tied to the usage, some will choose to self-moderate that wouldn’t if it were an outright ban. It sounds contradictory, but monitoring can produce better results than banning in some cases by avoiding fight “forbidden fruit effect” of using something because it’s off limits and gives the user autonomy to make the decision without it being forced up on them.
Giving AI controlled access to data it’s allowed to use
For a lot of organizations, a single approach won’t fully stop Shadow AI. And it’s not always obvious which will work best.
This is where Oveka comes in. Oveka allows you to see how AI is being used, gives AI access to data sets, and puts lightweight guardrails around what types of data can be accessed by users and their AI assistants. It lets you give AI controlled access to certain datasets and restrict the ways specific tools can be used.
Controlled AI access sits in the middle of banning and monitoring. It’s not an outright ban, but it’s not just monitoring. Users can embrace AI while sensitive data stays protected.
Think of it as a combination of both monitoring and managing how AI is used across the organization. First, you can see how AI is being used by identifying which tools are being used, by who, and in what way. Then you decide what happens next.
Monitor and control AI access to data sets with Oveka
Oveka shines a light on Shadow AI through two distinct angles.
PromptGuard monitors prompts and conversations and allows you to restrict specific tools from being used. It stops sensitive data from ever reaching unapproved tools.
Databridge integrates with SaaS platforms to give AI controlled access to specific datasets. It restricts what data AI can access and what types of data can be used in responses.
This combined approach means you can monitor, educate, and regulate AI use throughout the organization. And you always stay in control. Find tools, monitor usage, restrict what kinds of data can be used, and/or ban tools entirely, with granular controls to customize how it works for different tools, users, groups, or content types.
To find out more, get in touch with our team or book a demo now.