TL;DR: AI is already inside most organisations, whether through vendor tools, employee habits, or leadership pressure. The real issue now is not whether to allow it, but how to identify where it is, what it is touching, and how to govern it without slowing the business down.
If this piece resonated, share it with someone who is already trying to work out where AI is showing up inside their organisation. Thank you :)
Most organisations know AI is in the building. The harder question is what it’s actually touching, and whether anyone has a clear enough picture to answer that with confidence. That is where a lot of organisations are now.
It arrived through vendor updates, employee habits, and top-down encouragement. The tools people already use, the features vendors keep adding to platforms we already pay for, the habits people build at home and then quietly carry into work on Monday morning, and increasingly a push from leadership saying we need to do more with AI.
That is what makes this different.
It is not one decision. It is a series of small ones, made in different places, by different people, often without anyone stepping back to fully understand what the complete picture looks like.
It did not arrive in one clean rollout
A lot of enterprise technology still gets introduced through a fairly familiar process. Someone spots a need. A team evaluates tools. Security reviews happen. Legal gets involved. Procurement does its thing. Everyone eventually signs off and the platform gets rolled out.
AI hasn’t really respected that model.
Some of it is embedded in software we already use. Microsoft, Salesforce, ServiceNow, Canva, Workday, and others are all pushing AI deeper into their products. So what looks like a normal platform update can also be a change in how data is being processed, surfaced, or used.
Some of it is brought in by employees themselves. People try ChatGPT, Claude, Perplexity, Gemini, and the rest in their personal lives because they are fast, useful, mostly free and easy. Eventually, people start using the same approach at work. Not because they are trying to cut corners, but because it just feels normal.
And some of it is sanctioned from above. That is the part people do not always say out loud. When leaders start encouraging AI use without being equally clear about boundaries, they create a kind of cultural permission that can quickly get ahead of policy and governance.
So the issue is not really whether AI is in the organisation. It is.
The issue is whether anyone has a clear view of where it is, how it is being used, and what risk it is quietly introducing.
Policies are lagging the behaviour
Most organisations are not starting from zero here. They already have privacy policies, acceptable use policies, security awareness training, and vendor review processes. The problem is that people do not always recognise how those controls apply to AI.
The organisation tends to assume software adoption happens through formal channels, that people will wait for approval, and that the risk is visible enough to be managed in a straightforward way. That is not really how AI is moving through organisations.
AI is cheap to try, easy to access, often embedded in the tools people already use, and usually immediately useful. That makes it harder to govern. By the time a policy is written, reviewed, approved, and communicated, the behaviour may already be well established, or it may have moved on again.
What I’ve found in practice is that it happens in the flow of work. If a feature is visible and useful inside a tool someone already has open, they’ll use it. Nobody stops to ask whether they should or shouldn’t. They’re focused on finishing the task, not on the governance question the task might be raising. And what feels obvious to a security or IT team about data risk isn’t obvious to someone trying to get a report out before a deadline. The dots between “I used the AI summary feature” and “that data just moved somewhere” don’t get connected, because there was no reason to connect them at that moment.
Add to that the understanding gap around AI implementations. A lot of people still cannot clearly distinguish between a public AI tool, an enterprise-hosted AI tool, and a vendor feature that processes company data inside a larger platform. They may assume that if it lives inside the enterprise, it is automatically safe. It is not that simple.
The real question is what data is going in, who can access it, how it is retained, what it is connected to, and what settings the organisation has actually chosen. That is where the risk sits.
Private does not mean harmless
This is where enterprise AI and private cloud deployments come in.
And to be fair, they do help. They can reduce some risk by keeping data inside a controlled environment and limiting whether prompts or outputs are used for model training. That matters. It is a better model than throwing sensitive information into a consumer tool and hoping for the best.
But private does not mean harmless. A tool may have access to more data than it needs. A permission set may be too broad. A transcript may retain more than people realise. A user may assume the system is safer than it is. An output may be copied somewhere it should not have gone.
And beyond those, there are risks that are harder to see. A tool may be connected to data sources it should not be seeing. It may also infer things people did not explicitly put in. A spreadsheet might only show basic records, but the AI could still surface patterns that suggest something more sensitive. It can also connect the dots across systems in ways no individual user asked for. That is what makes this risk easier to miss than the ones above it.
The danger is not just that AI might leak data in some abstract sense. It is that it creates new ways for data to move, get surfaced, and be misused, often faster than the organisation’s controls were designed to handle.
What matters now
So the question is no longer whether AI should be allowed in the organisation. It’s already here.
The better question is how we identify where it already exists, what it is touching, and which uses are acceptable versus which ones are introducing risks we do not yet fully understand. That is a different problem from the one most teams were expecting.
Because this is not just a technology problem. It is also a governance problem, a leadership problem, and a people problem. If leadership wants more AI use, then leadership also needs to support the guardrails that make that use sensible. If employees are going to use tools because they are useful, then the organisation needs clearer boundaries around what can and cannot be put into them.
And if vendors are going to keep adding AI features into platforms, then the organisation needs to understand those features better and stay awake to what has changed, not just what was originally contracted for.
What to do next
The answer is not to block everything. That usually just pushes behaviour underground.
The better move is to make the safe path easier than the risky one. Be clear about which tools are approved. Be clear about which data types can never go into external or consumer AI tools. Be clear about what enterprise AI does and does not solve. And be honest that no policy will create zero risk in a world where the technology is changing this quickly.
The goal is controlled use and better visibility. It is about knowing where AI is already being used, deciding what good use looks like, and putting enough structure around it that people can still move fast without creating blind spots.
That is the real task now.
The question worth asking right now is not whether AI is allowed. It is whether you actually know where it already is.
~ Pete G