A few years ago, we often talked about Shadow IT and how organisations should deal with it. Employees using Dropbox instead of the file server. Teams who independently purchased SaaS solutions without involving IT. Marketing departments who rolled out entire applications before the security team even knew they existed.
We now know how that turned out.
Not by blocking everything. Not by preventing users from being innovative. But by offering safe alternatives, establishing governance structures and gaining an understanding of what was actually happening.
Today, we are actually seeing exactly the same pattern emerge.
It’s just that it’s much quicker now.
Shadow IT developed over a period of years. Shadow AI, on the other hand, spreads throughout organisations within weeks or even days. Not only via standalone applications, but via browsers, SaaS platforms, mobile apps and, increasingly, autonomous AI agents. It is precisely this speed that makes it a fundamentally different challenge.
Shadow AI and AI governance have now become much-discussed topics. Nevertheless, I believe that many organisations still significantly underestimate the scale of this development.
The claim: Everyone uses AI. We know that by now, but there’s a lack of clarity on which tools are used and how.
During conversations with clients, I often ask a simple question:
Which AI tools are used within your organisation?
This is usually followed by a neat answer. Microsoft Copilot. ChatGPT Enterprise. Perhaps an AI feature within a CRM or security platform.
Next, I’ll ask a second question: And which AI tools do your staff use?
Then there’s often a silence.
Research shows that a large proportion of staff use AI tools without explicit approval from IT or security. Gartner now cites ‘Shadow AI’ as one of the fastest-growing risk areas within information security. The challenge is not that employees are using AI, but that organisations often have no insight into which tools are being used, what data is being fed into them, and what decisions those tools subsequently make.
Incidentally, that behaviour on the part of your staff or colleagues is perfectly understandable.
Employees want to work more efficiently. They want to write reports faster, prepare presentations more quickly, generate scripts, and perform analyses with less effort. AI helps them achieve these goals, which is why many employees adopt AI tools before formal policies or approved solutions are in place.
Shadow AI is not a technical problem
Instinctively, we react by putting up defences such as
- Ban ChatGPT.
- Block Gemini.
- Enable a web filter.
Problem sorted. Right?
The reality is different.
AI is now everywhere. In browsers. In SaaS platforms. In CRM solutions. In development environments. In mobile apps. And, increasingly, in autonomous agents that carry out actions independently on behalf of users. Gartner now identifies three new blind spots: Shadow AI, Embedded AI and AI Browsing. Organisations can only manage AI risks if they first gain an understanding of all these forms.
Shadow AI is not, in essence, a technological problem. It is a governance problem.
The real question must be: What data is being entered?
That’s where the risk ultimately lies.
An employee who has a marketing text rewritten is not the same as an employee who uploads customer data, source code, contracts, financial information or strategic plans to a public AI service.
Recent research shows that data breaches involving generative AI are now among the biggest concerns for security teams. Various studies indicate that sensitive business information is increasingly finding its way into AI platforms without organisations being aware of it.
And that risk increases as AI becomes more autonomous.
Whereas we used to see mainly prompts, we now see agents gaining access to SharePoint, CRM systems, ERP platforms and document storage. As a result, the risk is shifting from the exchange of information to automated decision-making and automated actions.
From Shadow AI to AI Governance
I wrote earlier that, at the RSA Conference, Agentic AI wasundoubtedly one of the dominant themes. But the insights into governance were just as interesting.
Now that everyone is working with or experimenting with AI, there is a need for a mature approach to AI governance.
AI governance does not start with technology. AI governance starts with policy, risk assessment and ownership.
As CISO, there are three things I would do straight away. I certainly wouldn’t start with a block list, but as a first step: visibility
- Find out which AI services are actually being used. Not just the officially approved solutions, but also the browser tools, extensions, embedded AI features and autonomous agents that operate outside the IT department’s purview. After all, you can’t protect what you don’t know about.
- Next, establish clear rules. Not to stifle innovation, but to provide staff with a safe framework. What data is permitted to be used? What information is never permitted? Which tools are approved? What controls are in place?
- Investing in awareness is critical because most Shadow AI is not driven by malicious intent. Instead, it arises when employees seek more efficient ways of working, mirroring the patterns previously seen with Shadow IT.
From banning to making it safe
Perhaps that is the most important lesson of all. The organisations that successfully manage Shadow AI are those that offer secure alternatives; that combine governance with innovation; that combine transparency with trust; and that understand that AI has now become an integral part of their day-to-day work.
A good example comes from Check Point, for instance.It is no coincidence that, in a recent article , they talk about detecting, managing and protecting AI usage across browsers, SaaS platforms, desktop applications and AI agents. The focus is shifting from blocking to enabling in a controlled manner.
In conclusion
Shadow AI isn’t the problem. Invisible AI is the problem.
The organisations that will be successful in the coming years are not those that hold back AI for the longest. They are the organisations that know how to embrace AI in a responsible way.
And in the knowledge that innovation and security are not opposites, but in fact reinforce one another.
You can only act on what you see. And that starts with one simple question:
Do you actually know which AI is already being used within your organisation today?
Relevant resources from our partners:
https://www.trendmicro.com/explore/trendai-global-ai-study
https://www.checkpoint.com/ai-security/ai-workforce-security
Estimated reading time: 6 minutes