What does privacy-first mean for an AI assistant?

That memory is granted rather than assumed, that you can see exactly what is held, and that correcting, limiting, exporting or deleting it is a normal action rather than a support request.

Can I see everything the assistant remembers about me?

You should be able to. If a system cannot show you what it knows, you cannot meaningfully consent to it knowing anything.

What happens when I delete something?

It should stop influencing future behaviour, not merely disappear from a screen. Ask any provider that question directly.

Is my information used to train models?

Ask, and expect a clear answer with an opt-out. Guardian's published privacy notice is still in draft and under legal review; we would rather say that than overstate it.

Guide

Privacy-First Personal AI: Memory You Can Inspect and Remove

An assistant that is useful has to know things about you. That is precisely why control over what it knows cannot be an afterthought.

Privacy-first personal AI means the system's memory is something you grant, inspect and revoke rather than something it accumulates quietly. Guardian is designed so that people can inspect, correct, limit, delete and export what Guardian knows.

Start Guardian Daily Trust & governance

Opens the Guardian Daily app. You may need to sign in or create an account.

The trade

Usefulness requires memory — control makes it acceptable

An assistant with no memory asks the same questions forever and helps with nothing.
An assistant with unbounded memory becomes a record of your life you never agreed to.
The resolution is not less memory. It is memory that is visible, scoped and reversible.
Consent that cannot be withdrawn was never really consent.

The five controls

What you should be able to do at any time

Inspect

See what is held, in plain language, without filing a request.

Correct

Fix what is wrong. Bad memory produces confidently wrong suggestions.

Limit

Scope what may be remembered by category, source or time.

Delete and export

Remove it so it stops shaping behaviour, and take a copy with you.

Questions to ask

How to evaluate any personal AI product

Where is my data stored, and who inside the company can read it?
Is it used to train models, and can I opt out without losing the product?
What is the retention period, and what happens when I close my account?
Is there a record of what the system did on my behalf?
Is the privacy policy specific, or written to permit anything?

Where we are

What we can say honestly today

Design commitment

Memory by permission, inspection and deletion as first-class features, and no advertising model.

Still in draft

Our published privacy notice, terms and responsible-AI statement are drafts under legal review.

No overclaiming

We will not describe certifications, audits or guarantees we do not yet hold.

Questions

Common questions

What does privacy-first mean for an AI assistant?: That memory is granted rather than assumed, that you can see exactly what is held, and that correcting, limiting, exporting or deleting it is a normal action rather than a support request.
Can I see everything the assistant remembers about me?: You should be able to. If a system cannot show you what it knows, you cannot meaningfully consent to it knowing anything.
What happens when I delete something?: It should stop influencing future behaviour, not merely disappear from a screen. Ask any provider that question directly.
Is my information used to train models?: Ask, and expect a clear answer with an opt-out. Guardian's published privacy notice is still in draft and under legal review; we would rather say that than overstate it.

Keep reading

Related guides

More plain-language explanations of how Guardian Daily is designed.

You control what Guardian knows.

Inspect it, correct it, limit it, export it, delete it. That is the design, not a setting buried in a menu.