Building Trust: Our Approach to AI Privacy

Privacy isn't a settings page. In AI products, it's the foundation.

An AI assistant is only useful if it can work with real context -- inbox, calendar, documents, and communication history.

But that immediately raises the question: What happens to my data? Who controls it?

We take a privacy-first approach because trust is not optional in professional workflows.

Start with the basics: data minimisation and purpose limitation

If you're building in the UK or EU sphere, regulators are clear: collect what you need, and don't collect what you don't.

The UK ICO describes data minimisation as using the minimum personal data required for your purpose -- no more.¹ And GDPR principles emphasize purpose limitation: data should only be collected for specified, explicit purposes.²

This is more than compliance language. It is a product design principle.

AI adds new risks -- so we design for them explicitly

Modern AI security guidance repeatedly warns about:

prompt injection
insecure output handling
and sensitive information disclosure³

That's why we treat privacy as a system design problem, not just policy text.

What we mean by privacy as a user experience

A privacy-respecting assistant should feel:

predictable (no surprises)
visible (you can see what it's doing and why)
controllable (you decide what happens)

Approval-first plays a big role here because it reduces accidental or unwanted actions. It aligns with the broader human-in-the-loop approach used to improve accountability in automated systems.⁴

How the industry is talking about data use (and why it matters)

It is also helpful to know how major platforms describe their data practices:

OpenAI states that, by default, business data is not used for training in its enterprise offerings, unless you explicitly opt in.⁵
Google's Workspace privacy docs say chats and uploaded files in Gemini for Workspace are not reviewed by humans or used to train models without permission.⁶

We are citing these not because big company says so equals perfect privacy, but because it shows where the bar is moving: clear commitments and control.

The Clastines privacy mindset

Our north star is simple: Clastines should feel like a colleague you trust -- not a black box you tolerate.

That means designing around:

minimisation (only what's necessary)
purpose limitation (clear use boundaries)
guardrails against common AI risks
and approval-first execution for meaningful actions

Trust is the product. Privacy is how we earn it.

References