Is it safe to paste this into AI?

The team has figured it out—on their own. Sarah in accounts pastes the client fee schedule into a free chatbot to tidy it up. Mark pastes a survey report, client's home address and all, into another to rewrite the summary. Both use personal accounts.

None of it was malicious. But those conversations now sit with providers Mossvale has no contract with. Free AI apps can keep chats, let human reviewers read some of them, and be ordered by a court to preserve them—in 2025 a US court did exactly that for ChatGPT conversations, deleted ones included.

The work doesn't get any lighter, so the AI simply moves. Reports get drafted on personal phones over lunch and emailed back in. Nobody mentions it, because officially it isn't happening.

Now you can't see any of it. This is shadow AI, and it's costly: IBM found that breaches at organisations with high levels of shadow AI cost an average of US$670,000 more. Even Samsung's ban, after engineers pasted source code into ChatGPT, was a pause while it put proper controls in place—not the answer itself.

Reports are going out faster, using five different AI apps across twenty personal accounts. It feels like progress, until the client chases Question 14: which AI tools hold our data, and where?

You can't answer. If staff have put clients' personal details into those tools, UK GDPR still applies: Mossvale needs a lawful basis, and data sent to providers outside the UK brings the international-transfer rules into play. Right now you can't even say which providers have it.

The team switch off "Improve the model for everyone", and you answer Question 14: "Our data is not used to train AI." It feels done.

But that setting only covers training. The provider can still keep conversations for a period—OpenAI's business platform, for example, keeps abuse-monitoring logs for up to 30 days. On consumer apps people may read them: Google says some Gemini chats are reviewed by people, and reviewed chats are kept for up to three years. And in 2025 a US court ordered OpenAI to preserve ChatGPT conversations, deleted and temporary chats included.

Meanwhile the chats still sit in staff members' personal accounts. Anyone who gets into one of those accounts, or is sent a shared chat link, can read what was pasted.

You buy a business AI that works across your Microsoft 365 files. On day two, a junior surveyor asks it to "summarise this year's pay review"—and it does.

The AI did nothing wrong. Years ago the HR folder was shared with "everyone", and assistants like Microsoft 365 Copilot can show a user anything that user can already open. Untidy permissions that nobody noticed are now one question away.

Nobody knows what "careful" means. Is a client's name all right? A draft contract? A week later, someone pastes the office router's admin password into a chatbot to ask how to set up guest Wi-Fi.

They thought they were being careful: it was only the Wi-Fi. But a password pasted into an AI tool can't be taken back, so it has to be changed—and the contract pasted the day before is already out of your hands.

Taking names out is a good habit, and it works most of the time. Then report 37 goes in with a tenant's address in a photo caption, into a free tool on someone's personal account.

You still can't tell the housing association where their tenants' data is held, because it depends on whoever was typing that day.

The answers are better, but the accounts are still personal, the terms are the consumer ones, and there's no policy on what goes in.

When someone leaves, their chat history—tenant details and all—leaves with them. Paying for an AI app doesn't make it your business's AI.

Your answer to the housing association fits in a paragraph: which AI is used, where tenants' data is held, who can see it, how long it's kept, and what staff may and may not put in. Their data protection officer has no follow-up questions.

The team uses the AI openly because they know the rules. Reports go out faster, the project folder is locked to the people working on it, and nobody is drafting on their phone at lunch.

Two hundred reports, a tight deadline and a team that knows AI could halve the work. By week three, some of them are quietly using it anyway, on their own devices.

It's the ban again, at the worst possible moment: shadow AI on your most sensitive client's data.