A recruitment agency’s consultant needs to write shortlists quickly. She pastes ten candidates’ CVs, complete with names, ID numbers, contact details and salary history, into a free AI chatbot and asks it to compare them. It does a good job in seconds.
It doesn’t occur to her to ask where that information has gone, who can see it, or whether it might be used to train future versions of the tool. Nor does it occur to her that, under South Africa’s data protection law, the agency is responsible for how candidates’ personal information is processed and shared.
AI tools are often so easy to use that they feel like private notebooks. They aren’t always. Depending on the tool and the plan, what you type may be stored, reviewed by people or used to improve the service. For small businesses handling customers’ and employees’ information, understanding this is one of the most important parts of choosing and using AI tools.
One Like insight: Treat every prompt as if you’re sharing it with a third party, because sometimes you are.
When confidential data meets AI
One of the most widely reported early warnings came in 2023, when engineers at Samsung Electronics reportedly pasted confidential material, including source code and internal meeting notes, into ChatGPT while trying to solve work problems. Once the company realised that information entered into the public service was leaving its control, it restricted employees’ use of such tools and moved to develop internal alternatives. Several other large companies, particularly in finance, introduced similar restrictions around the same time.
The underlying issue is that different products handle data in different ways, and those differences are easy to overlook. Broadly, AI providers often distinguish between consumer versions, where conversations may be retained and, depending on settings, used to improve models, and business or enterprise versions, which typically offer stronger commitments, such as not training on customer data, data retention controls and contractual protections. Policies change regularly and vary by provider, so the only reliable approach is to read the current terms of the specific tool and plan you use.
For South African businesses, there’s also a legal dimension. The Protection of Personal Information Act governs how personal information is collected, used and shared, including with third-party service providers and, in some cases, across borders. Entering customers’ or employees’ personal details into an external AI service is a form of processing that needs to be handled within those rules, as we explored in POPIA Has Teeth Now.
Choosing tools with data protection in mind
A few practical steps help small businesses use AI without putting sensitive information at risk.
Know which plan you’re on. Check whether your AI tool is a free consumer version or a business plan, and what its current terms say about data retention and model training. If in doubt, assume consumer versions are not suitable for confidential information.
Check the settings. Many tools let you switch off chat history or opt out of your data being used for training. Make sure these are set appropriately for business use.
Anonymise where you can. Replace names, ID numbers, contact details and account numbers with placeholders before asking an AI to help analyse or draft something. Often the task works just as well without identifying details.
Keep the most sensitive information out entirely. Health records, financial account details, passwords, legal matters and anything under professional confidentiality usually don’t belong in general-purpose AI tools at all.
Write it into your AI guidelines. As we explored in Your Team Is Probably Already Using AI, a short written policy is the simplest way to make sure everyone follows the same rules.
[Infographic placement] Before you paste. A simple traffic-light chart. Green, generally fine: your own marketing copy, public information, anonymised examples. Amber, check your plan and anonymise: internal documents, client project details without names. Red, keep out of general AI tools: ID numbers, health records, bank details, passwords, confidential legal or HR matters. Caption: Data handling varies by provider and plan; always check current terms.
A quick data check
List the AI tools used in your business and, for each, find out three things: which plan you’re on, what the current terms say about how your data is stored and used, and which settings you’ve chosen. If a tool is used for anything involving customers’ or employees’ information, consider upgrading to a business plan with stronger protections, or using it only with anonymised information.
Then remind the team of a simple rule: if you wouldn’t email it to a stranger, don’t paste it into an AI tool unless you’re certain the tool and plan are approved for that kind of information.
A few questions can help:
- Which AI tools and plans does our business use?
- What do their current terms say about data retention and training?
- Has any personal or confidential information been entered into consumer AI tools?
- Do our staff know what must never be pasted into AI?
Which brings us back to the recruitment consultant and her ten CVs. The comparison saved her time, but it meant candidates’ personal details had been shared with an external service without anyone checking the terms. Using a business-grade tool with appropriate protections, or replacing names and ID numbers with placeholders first, would have delivered the same speed without the risk.
Sources
- Reporting on Samsung Electronics restricting generative AI use after confidential data was entered into ChatGPT, April–May 2023 (for example Bloomberg and TechCrunch).
- Protection of Personal Information Act 4 of 2013 (South Africa), general reference on processing personal information and transfers to third parties.
