The short answer
Public AI chat tools — ChatGPT and the many tools like it — are very good at writing, summarising and explaining. Whether it is safe to put company data into one depends on three things: which tool and which plan (free personal accounts and business plans often have different data terms), what data (a press release is not a payroll file), and what happens around it (who can see it, whether there is a record, whether it is used to train the tool).
So the honest answer is: sometimes yes, often not without rules. Each vendor's terms change over time, so check the current terms of the exact tool and plan your team uses — not a blog post, including this one.
The real risks, in plain words
1. Copy-paste leaks
The most common risk is not hacking. It is an employee pasting a customer list, a price file or a salary table into a personal account to "just get a quick summary". The data has now left the company, and nobody knows it did.
2. No permissions
Inside the company, a sales rep does not see payroll and a warehouse worker does not see margins. A public chat tool does not know your company’s permissions. Whatever is pasted in is out of your access controls.
3. No connection to live systems
Unless it is connected to your systems, a public chat tool does not know your ERP. It sees only what someone pasted, from whenever they exported it. Yesterday's spreadsheet becomes today's answer.
4. Answers not grounded in your numbers
Ask a general AI tool about "our margins" and it may produce a fluent, confident answer built on partial data or on assumptions. For a decision about money, an answer has to come from your own systems, not from a plausible guess.
5. No record
When questions and answers live in personal accounts, there is no company log of who asked what, what data was used and what came back. That matters for security, for privacy obligations around customer and employee data, and simply for checking a number later.
What to check before approving any AI tool
Use these questions with any AI vendor — a public chat tool, an assistant built into software you already use, or WizBI. The right column shows WizBI's answers.
| Question | Why it matters | WizBI's answer |
|---|---|---|
| Where does our data live, and who processes it? | You need to know which cloud and which companies touch your data | Runs on Google Cloud; each organisation's data is isolated and encrypted in transit and at rest |
| Is our data used to train AI models? | Data used for training can be hard to take back | AI models run inside Google Cloud under enterprise terms; customer data is not used to train them |
| Who can see what? | An AI tool should not widen access beyond your own rules | Permissions by organisation, group and user — the same person has the same permissions on every channel |
| Is there a log? | You must be able to check who asked what, and what came back | Every question is logged and reviewable |
| What can it do in our systems, and who decides? | An AI tool should act only within rules you set | Access is agreed with you per system; any automation is defined with you, runs under your permissions and every action is logged |
| Is the vendor certified? | A certification means an outside auditor checks the controls | WIZBI AI LTD is ISO/IEC 27001:2022 certified |
| Are answers based on our data? | Decisions need your numbers, not a plausible guess | Connected to your systems; answers come from your own data |
A practical AI policy for your team
Banning AI tools rarely works — people use them on their phones anyway. A short, clear policy works better:
- Sort data into three groups. Public (fine anywhere), internal (only in approved tools), and sensitive — customers, employees, prices, contracts, financials (only in tools with permissions and a log).
- Approve tools by plan, not by brand. The same product can have very different terms on a personal account and on a business plan.
- Give people a better option. Most copy-paste happens because someone needs a number from the ERP and has no other way to get it quickly. Give them one.
- Keep a record. Whatever tool you approve for company data should log questions and answers.
- Review every few months. Tools and terms change. So does what your team uses.
A different setup: ask the company brain instead of pasting
WizBI turns the problem around. Instead of copying data out of your systems into a chat tool, it connects to the systems themselves — Priority, SAP, monday.com, SQL databases, Excel and Google Sheets and more — and builds one company brain that knows the business, its terms and its permissions. Your team asks in WhatsApp, Microsoft Teams, Slack or the web, and each person gets an answer only from what they are allowed to see.
If your team prefers an AI assistant such as Claude or Microsoft Copilot, it can connect to WizBI via MCP — the assistant then gets answers from the company brain, with the same person's permissions. See MCP for how that works, and SQL databases for connecting an on-premise database through a secure agent with no inbound ports.
Where public AI chat tools are still the right tool
Public AI chat tools are excellent for work that does not need your private data: drafting an e-mail, rewriting a paragraph, explaining a term, brainstorming a presentation. Keep using them for that. The line is simple: general work in general tools; questions about your company's data in a tool that is connected to it, with permissions and a log.