Quick answer: To use AI safely at work, keep personal data and confidential company information out of public chatbots, check every output before you rely on it, and use your employer’s approved tool and settings where one exists. Government cyber agencies in the UK and Canada give the same core advice.
Most workplace AI trouble comes from three things: pasting in something you shouldn’t, trusting an answer that’s wrong, and using a tool your employer hasn’t approved. None needs technical skill to avoid.
This guide gives you a simple routine you can apply to any chatbot. For the wider picture of tools and news, see our AI news and guides.
What should you never paste into an AI chatbot?
The UK’s National Cyber Security Centre (NCSC) says that queries sent to a public AI service are visible to the company running it, and may be used to develop the model. Its advice is to avoid including sensitive information, or anything that would cause problems if it became public. See the NCSC blog post.
Canada’s Cyber Centre says the same, advising users to avoid providing personally identifiable information or sensitive corporate data to these tools. Read its generative AI guidance.
In practice, keep these out of public chatbots:
- Customer or colleague names, emails, addresses and account details.
- Passwords, API keys and anything from your password manager.
- Unreleased financials, contracts, legal advice and board material.
- Source code or internal documents your employer treats as confidential.
- Health, HR or disciplinary information about any person.
A useful test: would you be comfortable seeing this text on a stranger’s screen? If not, don’t paste it.
How do you get the benefit without sharing the data?
You don’t have to give the chatbot the real thing. Replace names with roles (“Client A”), swap exact figures for round or invented ones, and describe the document instead of uploading it. Ask for a template or structure, then fill in the sensitive parts yourself.
Which settings protect your chats?
Settings vary by tool and plan. For ChatGPT, OpenAI’s help center says you can turn off “Improve the model for everyone” under Settings, then Data controls, and that Temporary Chats aren’t used to improve its models. It also says business and enterprise workspaces aren’t used for training by default. Giving a thumbs up or down can still cause that conversation to be used. Details are in OpenAI’s data controls FAQ.
The wider point: a work account with admin controls is usually safer than a personal one. If your company provides one, use it. Our guide on how to stop AI chatbots training on your chats walks through the settings across the main tools.
How do you check AI output before you use it?
Canada’s Cyber Centre warns that AI output can be incorrect, nonsensical or biased, and recommends fact-checking against credible sources. The NCSC similarly describes chatbot answers as convincing but sometimes only partially correct.
- Check numbers, names and dates against the original source.
- Click every citation. If a link or quote doesn’t exist, assume it’s made up.
- Ask for the reasoning, then check whether it holds up.
- Get a second pair of eyes for anything legal, medical, financial or going to a client.
- Take responsibility. If your name goes on it, the checking is on you, not the tool.
What should you ask your employer?
Before relying on AI daily, find out whether there’s an AI policy and which tools are approved. If you handle personal data in the UK, the Information Commissioner’s Office says organizations should think about data protection impact assessments and transparency when using AI, though it notes its guidance is under review after recent law changes. See the ICO’s AI and data protection guidance. In the EU, new rules on AI are also arriving in stages; our piece on https://insightorys.com/2026/10/07/eu-ai-act-labels-2026/ explains what applies now.
If there’s no policy, ask for one in writing. If you’re self-employed or run a small business, write a one-page version for yourself: allowed tools, banned data, and who checks outputs.
A five-minute routine before you hit enter
- Ask what the task needs. Most prompts need a topic and a goal, not a customer’s details.
- Strip the identifying bits. Names, numbers, addresses and anything that points to one person.
- Choose the right account. Use the work-approved tool for work, and a personal account only for non-sensitive tasks.
- Read the answer line by line. Skim-reading is how errors reach clients.
- Save only what you need. Delete chats you don’t need, and avoid storing confidential material in memory features.
What about AI notetakers and browser add-ons?
Meeting recorders and AI browser extensions can see far more than a chat box does, including everything said in a call or shown on a page. Ask your IT team before installing one, tell meeting participants when you’re recording, and avoid running them on calls that cover personal or confidential matters.
Where AI is a good fit at work
Low-risk uses include brainstorming subject lines, summarizing a public article, tidying your own notes, drafting a polite reply to a generic request, or explaining a spreadsheet formula. These tasks don’t need private data, and mistakes are easy to spot.
Higher-risk uses include anything involving a named person, a legal commitment, a medical question or a financial decision. Use AI to prepare, then let a qualified human decide. If you aren’t sure which bucket a task falls in, treat it as higher risk.
Don’t forget the basics
AI also makes scams more convincing. Canada’s Cyber Centre points to multi-factor authentication, software updates and caution with unexpected links and attachments. Phishing emails written by AI can look polished, so judge a message by what it asks you to do, not by how well it’s written.
Frequently asked questions
Can my employer see what I type into ChatGPT?
On a personal account, no, but the AI company can. On a company-managed workspace, administrators may have access to logs depending on settings and your employer’s policy, so check before assuming privacy.
Is it safe to upload a work document to an AI tool?
Only if your employer has approved the tool for that kind of document. With a public chatbot, avoid uploading anything confidential or containing personal data.
Does turning off chat training make my chats private?
Not completely. It stops the setting from feeding model improvement, but the provider still handles your data, and feedback ratings can override the choice in some cases.
Do I have to tell people I used AI?
Rules vary. Some employers, schools and publishers require disclosure, and in the EU certain AI-generated content must be labelled. When in doubt, say so.