top of page

Bring Your Own AI: The Silent Risk of Staff Using Personal ChatGPT and Claude Accounts at Work

5 days ago
5 min read

Right now, somewhere in most businesses, an employee is pasting something into a personal ChatGPT or Claude account to get a job done faster. A client email to redraft. A contract to summarise. Some financial figures to tidy into a table. A half-formed product plan to sharpen up. They are not being reckless or disloyal. They are using a genuinely useful tool, on their own subscription, to do their work well. The problem is that their employer has no idea it is happening, no control over where that information goes, and no way to get it back once it is gone.


Woman working on a laptop in an office, resting her chin on her hand as a chat-style screen glows beside a phone.

This has a name now: Bring Your Own AI, or BYOAI. It is one of the fastest-growing risks facing businesses, and almost no small-business guidance covers it. What it is, why it is a different and trickier problem than ordinary unapproved software, what the real dangers are, and how to respond without simply banning tools your people find valuable are all worth setting out plainly.


What Is “Bring Your Own AI”?

It is staff using their own, personally-paid-for AI tools for work tasks, outside any oversight from their employer. And it is not a fringe behaviour, it is close to universal. Microsoft’s research found that 78% of people who use AI at work bring their own tools to do it, rising to 80% at small and medium-sized companies, where formal AI programmes are rarest. Gartner put the share of employees using personal AI accounts for work at 57%, with a third admitting they had uploaded sensitive company information to tools their employer never approved. Separate research found that 77% of employees have pasted company information into AI tools, and 82% of them did so through personal accounts rather than managed business ones. People are even paying for it themselves: around a third buy AI tools out of their own pocket to use at work, which tells you they will keep using them whatever the official line is.


The reason is simple and sympathetic. The tools help, the business often has not offered an approved alternative, so people quietly reach for their own. The demand is real; the oversight is missing.


Why Is This Different From Normal Shadow IT?

Because the usual ways of spotting and controlling unapproved technology do not work on it, and because what gets lost cannot be recovered. Traditional shadow IT, someone installing an unsanctioned app or signing up to an unapproved service, at least leaves traces: software to detect, network traffic to see, an account to find. BYOAI leaves almost none. It is a browser tab and a paste. Nothing is installed, nothing is procured, and the activity is largely invisible to anyone looking, because it runs through a personal account entirely outside the company’s systems.


The bigger difference is what happens to the data. When an employee pastes a contract or a customer list into a personal AI account, that content can be absorbed into the tool’s training, retained indefinitely, and potentially surface in fragments to completely unrelated users later on. You cannot un-paste it. With ordinary shadow IT, you can usually revoke access and recover or delete the data. With BYOAI, the information may already be gone for good, baked into a system you have no relationship with. There is a departing-employee twist too: because the account is personal, when the employee leaves, everything they fed into it goes with them, still sitting in their private account. And the newest wrinkle, personal AI agents that can connect into company systems through an individual’s own keys, creates exposure that does not resemble old shadow IT at all. It is shadow IT that is harder to see and impossible to take back.


What Are the Real Risks?

Woman in an office holds a smartphone and uses a laptop with blue charts, looking focused at her desk beside a plant.

They fall into three buckets, and all three are serious. The first is data leakage. Client communications, financial data, contracts and plans flow into tools the business cannot see or control, and the share of pasted content that is genuinely sensitive has tripled in two years. The second is loss of intellectual property. Source code is the single most leaked category, and strategic plans, pricing and proprietary methods pasted into a public tool can become training data that, in effect, leaves your competitive advantage sitting in a system others can query. The third is data protection and GDPR exposure. Your business remains legally responsible for the personal data your staff handle, and feeding customer information into an uncontrolled third-party AI tool, with no agreement in place about how it is stored or used, is exactly the kind of uncontrolled processing that creates compliance risk. On top of all that sits plain security risk: a compromised personal AI account becomes a quiet route to everything that has been pasted into it, and shadow AI is already turning up in real breaches, adding substantially to their cost. This builds on the wider problem we covered in our piece on how AI tools are changing where your company’s data goes.


Why Banning Doesn’t Work

The instinct is to forbid it, and it is the response most likely to fail. People use these tools because they save real time and genuinely help, and many care enough to pay for them personally. Tell them no, and they do not stop; they simply move the activity further out of sight, onto their phone or a home device, where you have even less visibility and the business loses the productivity into the bargain. A ban also ignores the actual cause. The reason staff reach for personal accounts is that most employers have not given them a sanctioned alternative, so they fill the vacuum themselves. Prohibition without provision just drives a useful, widespread behaviour underground. The goal is not to stop people using AI. It is to give them a safe way to do what they are clearly going to do anyway.


What Does a Sensible Response Look Like?

Three things, working together. First, provide a sanctioned, business-grade AI option, so people have a good tool they are allowed to use and no longer need their personal one. Business and enterprise tiers of these tools come with proper protections, typically not training on your data and offering contractual safeguards, which removes the core danger in one move. Second, set a clear, simple policy on what information may and may not go into which tools, so staff know the rules rather than guessing, which we lay out in detail in our piece on the AI policy every UK business now needs. Third, run approved AI inside a controlled, managed environment, so the data stays on home ground and the business keeps visibility rather than losing it to a scatter of personal accounts.


Put plainly, redirect the demand rather than forbidding it. Give people a sanctioned tool, tell them clearly how to use it, and provide a safe place to do so.


Why a Controlled Environment Helps

The cleanest way to hold all of this together is to deliver AI through a managed setup rather than leaving people to the open web. When approved AI tools run inside a controlled, UK-hosted environment, the data people feed in stays within a boundary the business governs, with the protections and oversight that personal accounts entirely lack. That is exactly how SystemsCloud builds AI into its AI-powered virtual desktops, and how our AI consultancy helps businesses choose the right sanctioned tools and put sensible guardrails around them. People get the speed they were chasing with their personal subscriptions, and the company keeps control of its information.


Redirect, Don’t Forbid

Bring Your Own AI is already happening across your business, almost certainly more than you realise, and the data walking out through personal ChatGPT and Claude accounts cannot be recalled once it has gone. It is a quieter and stickier problem than ordinary shadow IT, because it is nearly invisible and largely irreversible. Banning it only pushes it further from view. The businesses that handle this well are the ones that meet the demand head-on: a proper sanctioned tool, a clear policy, and a controlled place to use it. Do that, and you turn a silent risk into a managed advantage, with your people getting the help they want and your data staying where it belongs.

Comments


Contact Us

Thanks for submitting!

Have a question you want answered quicker?

Give us a ring or try our online chat!

Tel. 02039064600

Please do not block Caller ID so our team can assist you faster.

  • LinkedIn
  • Facebook
  • Instagram
  • Twitter

© 2026 SystemsCloud Group Ltd.

bottom of page