Why do staff use personal AI accounts for work?
The National AI Centre says that, whether or not an organisation has an approved policy, it is likely some staff are already using AI at work without their employer knowing. It calls this shadow AI and describes it as often a signal of unmet need, time pressure or curiosity from early adopters. People may also keep it quiet because they think using AI could be seen as cheating or make them look lazy or less capable. Read that way, personal use shows where the work pressure is and which tasks people think AI helps with, which is useful information for planning training.
What is the risk with a personal account?
The risk sits in the account terms and in what goes in. OpenAI's help centre says that for its individual Free, Plus and Pro plans it may use content to train its models unless the user turns off Improve the model for everyone in Settings, Data controls, and that feedback on a response can release the whole conversation for training even after an opt-out. The organisation also has no administrator control over a personal account: it cannot set retention, see what was shared or remove access when someone leaves.
- Client, patient or employee details typed into a consumer chatbot.
- Confidential documents uploaded for summarising.
- Work history that leaves with the employee.
- No shared rule, so each person decides what is acceptable.
How can you find out what is happening without blame?
Ask in a way that makes honest answers safe. A short anonymous survey works: which AI tools people use for work, on which tasks, on what kind of account, and what types of information they have entered. Say in advance that the purpose is to set rules and provide a proper tool, and that nobody will be disciplined for past use. Managers who describe their own use first make it easier for others. The National AI Centre advises creating a safe space, because people need to feel safe to disclose their use of AI before they will tell you about it.
What should staff be given instead?
An approved tool on work accounts, with a rule on what information may go into it. The tool is usually the one attached to systems you already pay for, such as Copilot Chat in Microsoft 365, Gemini in Google Workspace or a ChatGPT Business workspace. For ChatGPT Business and Enterprise, OpenAI says it does not use inputs or outputs to improve its models by default. The rule should fit on one page: which tool, which kinds of information never go in, what must be checked before an output is used, and who to ask when unsure.
What if someone has already entered personal information?
Deal with it calmly and specifically. Record which tool and account were used, roughly when, and what type of information was entered, without asking the person to paste it anywhere else. Check whether the conversation can be deleted and whether the account's training setting was on. OpenAI's wording is that after an opt-out it will not use new conversations, so earlier conversations are a separate question. Then take it to whoever handles privacy in your organisation, who can decide what follow-up your obligations require. This is administrative guidance, not legal advice.
Where does training fit in?
Training turns the rule into a habit. The OAIC says that any controls or prohibitions on including personal information in AI inputs need to be accompanied by training and auditing measures. A session for staff who have been using personal accounts should start from the tasks they told you about, show the approved tool doing those tasks, and practise the rule on realistic fictional material. Self-taught users vary a lot in how carefully they check outputs, so spend time on reviewing outputs against the source as well as on writing prompts.