AI Data Privacy

This data was originally featured in the April 26th, 2025 newsletter found here: INBOX INSIGHTS, April 23, 2025: AI Integration Strategy Part 3, AI Data Privacy

In this week’s Data Diaries, let’s talk about AI data privacy. I recently had the opportunity to speak at the Trace One User Conference in Miami, and one of the top questions on everyone’s minds around AI was data privacy. When is your data private? When is it not?

To clarify this, let’s first set some basic definitions. An AI model is software written by AI, for AI, in the same way that Microsoft Word is software written by humans, for humans. When we talk about AI models, we’re talking about the engines that do the data processing, converting prompts into outputs.

No one ever uses an AI model directly; they’re basically giant databases of statistics, if you were to open one up. Instead, we have interfaces to those models, tools like ChatGPT that talk to a model. You’ve seen this whenever you tap on the model picker in ChatGPT or Gemini and receive a bewildering, poorly-named list of choices – o3, o4-mini, o4 mini-high, GPT-4.5, GPT-4.1, etc. Those are the models, and ChatGPT is the interface.

This is important to know because, from a data privacy perspective, the interface is where privacy issues occur. The models themselves can’t and don’t ever learn directly from us. They’re massive pieces of software that take months to build and train. Like the engine of a car, we don’t interact with it directly. We interact with a steering wheel, seats, tires, the radio, etc.

What AI providers do, depending on privacy policies, is capture the data that you put into their interface and capture the data that comes out of the interface. That’s where you can lose the privacy of your data.

This is important, especially for understanding the difference between a model and an interface. DeepSeek, for example, is an interface on the company’s models. If you download their models (which you can) and run them on your own hardware, you have to provide the rest of the car – the interface. The same is true of every open weights model – you download their models engine, and then you put it in a car you have to provide.

That means that you control how data is collected and used. Your data is as safe as the rest of your infrastructure, be it a laptop, a server, or a huge, expensive AI cluster.

When you use someone else’s interface, you are subject to their privacy policies. DeepSeek’s web interface and app state clearly in their Terms of Service that they capture every bit of data that goes in and out of their model when it’s run in their interface. You get in their car, and they track everything you do.

The same is true of the free versions of most AI tools. The classic rule applies:

If you’re not paying, you (and your data) are the product that the company sells to someone else.

If your employees are using AI at work that you’re not paying for, they’re almost certainly leaking your confidential data to someone else. The solution for this is straightforward: pay for AI tools for your employees to use, and apply the same data governance and data security policies to those tools that you already have.

What’s not effective? Telling employees they can’t use AI at all. First, some employees will anyway, whether you want them to or not. Second, in doing so, you handicap your company’s capabilities compared to competitors that have strong AI policies and provide approved tools to their employees.

Think back to the early 2000s, when companies needed to figure out smartphones. Those companies that outright banned them had less satisfied employees who used them anyway, against company rules, and your data found its way onto devices outside your control. Those companies that figured out device management and paid for company devices for employees to use found greater adoption, greater satisfaction, and greater data security.

That’s where AI is today. Know what the privacy policies of different AI services are, and provide your employees with the tools and security that meet both your needs.

Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.