Why Your Business AI Tool Might Be Putting Confidential Conversations at Risk

It’s amazing how quickly we’ve become used to using tools like ChatGPT and Gemini on a daily basis. Whether you’re a fan of the LLM or you’re on the fence about how much you actually want to be striking up conversations with it, there’s no denying that for work and much of our personal lives, it’s found a use case. 

But we also know, to varying degrees, that some of these LLMs are a little iffy when it comes to privacy. Claims of platforms utilising and storing user data – even after chats are deleted – for training purposes have been swirling for a few years now, and there has been a stark rise in the number of warnings being sent out to businesses and individuals who might be getting a little too comfortable with what they’re willing to share. 

Here’s what you need to know. 

What’s the Risk? 

The risk begins in the design itself. LLMs are conversational. They don’t feel like the depthless chatbots we’re used to communicating with on e-commerce sites when we’re trying to figure out why our orders are taking so long. They’re capable of offering ‘insight ’-or mimicking it very convincingly - as well as reassurance and patience, and asking questions that keep the back-and-forth going. There’s nothing inherently wrong with that, and it’s a great way to make complex topics accessible – particularly in a busy workplace – but it does mean that users lower their guard pretty quickly. 

This is where sensitive information can quickly start to leave an organisation and become shared knowledge with whatever platform your employee is using. It may not be a deliberate decision to spill the company’s biggest secrets. It’s probably just a consequence of asking for help with Task A, or getting the LLM to check the employee’s work on Task B. 

Client proposals, pricing documents, internal strategies, and HR correspondence – these are all things you wouldn’t just send to another person or business without a strict NDA in place, but many people are letting them go straight into the LLM’s memory reserves without always knowing how deep they run.

Google searches are generally broad and generic, but LLM conversations often involve a lot of information, context, and insight that never used to go beyond company walls. 

The trouble is, the default settings on many free-tier AI tools permit data retention for the purposes of model improvement. It’s there in the terms and conditions, but most of us don’t stop to read them. Even if it’s not used for training, data is often stored or even viewed by human AI trainers.

What’s the Solution? 

There’s no need to decree that all AI tools are suddenly banned. There really is no space for that in modern business, and leaning too hard into the fear will only cause you to be left behind. 

The most reasonable alternative is, when you stop to think about it, the most obvious: find an AI assistant for business that prioritises security, safety, and privacy for all of its users, rather than masking data retention and usage in lengthy terms and conditions. Just like any other business tool, there’s a spectrum of options on the market that range from unreliable and untrustworthy to privacy-led and built-for-business.

An AI tool that is safe to rely on and share sensitive information won’t use data for training, won’t log chats, and will offer transparency via open-source code – among other things. It will be built around serving your business, rather than using its data for an ulterior purpose.