AI

AI vs privacy: How safe is your data?

  • AI is powerful, but are you giving away more data than you realise?
  • From ChatGPT to workplace AI tools, learn what happens to the information you share.
  • Simple privacy tips to help you use AI with confidence and keep sensitive data secure.

We’ve entered an era where AI tools like ChatGPT and Microsoft Copilot are as commonplace as email. These days, it’s not unusual for students to write essays with AI tools or for professionals to use them to summarise meetings. Large language models - and similar models that can process different types of data beyond text, such as images, audio, and sometimes even video (known as multimodal models) - have found their way into everyday routines. 

Considering how commonplace the use of AI tools has become, their privacy should be a pressing concern for individuals, and even more so for businesses. 

The invisible data trail 

Every time you type a prompt, upload an image, or share a document with an AI assistant, you’re creating a digital breadcrumb trail. Often, you might be sharing more than you realise. 

Take an seemingly innocent example: say you upload a photo of your fridge to ask for recipe suggestions. That image reveals your dietary preferences, favourite brands; it even hints at your income bracket or household size. 

Even though this is unstructured data, it's still DATA. And the right AI model, it can be interpreted, categorized, and potentially be used. 

What happens to your data? 

OpenAI, the company behind ChatGPT, has stated that they ‘do not share user content with third parties for marketing purposes’. However, they do work with a number of subprocessors: third-party companies that provide services like moderation, data storage, and infrastructure support. 

Unless you opt out, your interactions with ChatGPT (outside of API, Enterprise, or Team plans) may be used to improve their models. This includes analysing prompts and responses to refine how these models understand and generates content. 

And remember, it's not just about language - interactions with models like the ones behind ChatGPT can also be used to improve how they handle different types of data and help them evolve. And this can be text, images, audio, and so on. 

In other words, your data might not be sold - but it could still be stored, studied, and used to train future versions of the model. While OpenAI offers features like “Temporary Chats” and data deletion requests, the default settings still lean toward data retention. 

Regulatory frameworks are evolving 

Recent legal developments have added complexity.  

A U.S. court order in 2025 required OpenAI to preserve all ChatGPT logs, including deleted chats, as part of ongoing litigation. The case in question was brought by the New York Times and other media outlets, and was all about retaining the evidence of users potentially prompting AI to reproduce copyrighted material. 

OpenAI pushed back, citing user privacy concerns - but the case highlights how quickly the rules around AI and data can change, and how fragile our assumptions about privacy may be. 

The outcome: the order stands, and now affects users of ChatGPT Free, Plus, Pro, and API services. However, it does not apply to ChatGPT Enterprise, ChatGPT Edu, or users with a Zero Data Retention agreement  

OpenAI is appealing the decision, arguing that it undermines user privacy and may conflict with international data protection laws like the GDPR . 

For users, the reality is that, at least for now, even if you delete your chats, some version of that data is still stored somewhere. 

So, what should we be asking? 

The real issue isn’t just whether AI companies are acting responsibly or whether all their sub-processors are secure enough - it’s also whether users fully understand what they’re sharing in the first place. 

Should there be stricter data protection laws for AI interactions? Possibly. 

Should AI companies be transparent about how data is used? Absolutely. 

But perhaps most importantly: should users be more mindful of what they input into these tools? Yes – and urgently. 

We need to start treating AI prompts the same way we treat social media posts or emails - because they can be just as revealing, if not more so. That means thinking twice before uploading personal photos, sharing sensitive documents, or even describing private situations in detail. 

A call for AI literacy 

As AI becomes more embedded in our lives, we need a new kind of digital literacy - AI literacy. 

This includes understanding how AI models work, what data they collect, and how that data might be used. It also means recognising that even unstructured data - like a casual chat or a blurry photo - can contain a surprising amount of information about you, and more! 

In the age of ubiquitous AI use, privacy isn’t just about what companies do with your data. It’s about what you choose to share in the first place.

Explore our AI courses and skills training. 

Related articles

Deidré Luzmore explains why effective AI governance starts with leadership, accountability, and clear organisational oversight.

AI governance can't wait for regulation. Leaders can't either

6 November 2026

Cyber Portfolio Director Richard Beck shares practical insights on AI governance, assurance, accountability, and the risks of autonomous AI agents.

Expert spotlight: Richard Beck on AI governance, assurance, and accountability

6 November 2026

As AI moves from experimentation to widespread adoption, many organisations are discovering that governance is struggling to keep pace. While attention often focuses on future regulations, today's challenge is ensuring employees, leaders, and processes are aligned around responsible AI use.

As AI adoption accelerates, governance lags behind

4 November 2026

An unauthorised AI collective, a real-world cyber intrusion, and the uncomfortable question of whether human oversight and AI governance can keep pace.

When AI agents started talking to each other

26 October 2026

As AI becomes more capable and autonomous, who has the evidence, independence, and authority to decide when the risk is no longer acceptable?

Who decides when AI is too dangerous?

24 September 2026

David Pool explains how to build a cost-efficient AI enterprise by balancing model, governance, data, FinOps, and workforce capability.

Building the cost-efficient AI enterprise

10 September 2026

Why healthcare organisations must prioritise workforce transformation alongside AI adoption to achieve sustainable and effective change.

Before AI transformation comes workforce transformation

8 September 2026

Anthropic has announced a research preview of its Model Hardware Standard (MHS), a new specification designed to help AI agents interact with physical devices. If Model Context Protocol (MCP) gave AI a common way to discover and use software tools, MHS does something similar for hardware.

Anthropic's Model Hardware Standard: what happens when AI agents can use physical tools?

28 August 2026

Forward deployed engineers are emerging as a key AI role, helping organisations bridge the gap between technical capability and business value.

What the rise of the forward deployed engineer tells us about how organisations are building capability

27 July 2026

London’s plans to prepare workers for AI-driven change reflect what QA’s analysis of AI readiness across UK sectors has found: adoption is accelerating, but workforce skills, confidence, and capability aren’t keeping pace. By examining how AI is being adopted across industries, we’ve identified a widening gap between giving people access to AI and equipping them to use it effectively. 

Closing the AI readiness gap starts with skills

21 July 2026

About the Author

Ana Rojo Echeburua

Ana is a data and AI specialist with a strong background in mathematics who is very passionate about helping others solve real-world problems with them.
More about the author