Can You Trust an AI Assistant? Why Ranya Asks First
Help is only useful if it never surprises you. The difference between an assistant that reads and one that changes things, and why Ranya asks first.
By Shekib AzizPublished Updated 7 min read
You ask an assistant to clear out your inbox, and it archives the email from your accountant that you were waiting on. So can you trust an AI assistant? Not on its word.
You want help, not surprises. You can trust one that shows you each change before it makes it, then waits for you. That rule matters more than how clever the assistant is, and Ranya is built on it.
Two lists. Runs on its own: looking things up, remembering what you told her, opening a screen, and the seasonal effect on this device. Waits for your yes: every other change, including tasks, habits and goals, notes, email and sending it, focus (adding a task to one, starting, pausing or ending it), the calendar and settings.
Is it safe to let an AI assistant change your tasks, calendar or email?
It is as safe as its rule for changes. An assistant that answers a question wrongly costs you a moment: you hear the wrong answer and ask again. An assistant that changes something wrongly costs you more. A task you can't find, an email that went out, a meeting that moved.
And mistakes will happen. Speech gets misheard. People talk loosely ("the landlord thing", "move it to later"), and a model fills the gaps with its best guess. An AI agent's mistakes don't have to be dramatic to cost you. A small guess, acted on without asking, is enough.
Acting without asking is the real risk, and security guidance for AI systems names it. OWASP's entry on excessive agency gives the example of an extension that can delete a user's documents and does so without any confirmation from the user. Its fix is plain:
Require user approval
Utilise human-in-the-loop control to require a human to approve high-impact actions before they are taken.
So is it safe to give AI access to your email? It is when nothing it changes happens without your say.
Why do so many people distrust AI?
Most of us have been surprised by software before, and an assistant that acts for you raises the stakes. A global study by the University of Melbourne and KPMG asked people around the world how far they trust AI systems.
46%
only 46% of people globally are willing to trust AI systems.
In the United States, a 2025 Pew Research Center survey of adults found that half are more concerned than excited about AI in daily life, up from 37% in 2021. About six in ten would like more control over how AI is used.
The same survey has a second finding worth holding onto.
Nearly 3 in 4
nearly three-quarters say they'd be willing to let AI assist at least a little with day-to-day tasks and activities.
What does "human in the loop" mean?
It means a person approves an important action before an automated system carries it out. The system can prepare, suggest and look things up on its own. The step that changes something waits for a human. For an assistant, the line runs between reading and changing.
The people who build AI agents describe it the same way. Anthropic's framework for trustworthy agents says people should keep control over how their goals are pursued, particularly before high-stakes decisions are made.
What Ranya does on its own, and what it always asks first
Ranya draws the line in the same place, as the lists at the top show. Anything that only reads runs on its own. A change is read back first: Ranya says what she is about to do and waits for your "yes" or a tap on Confirm. "No" means nothing happens.
- You say it
- She reads it back
- You decide
- YesIt is done
- NoNothing changes
Diagram: you say it, she reads it back, you decide. Then it forks: yes, and it is done; no, and nothing changes.
Three small things run on their own, because each is easy to undo:
- A note she saves about you. You can read, edit or delete each one in Settings > Memory, and how her memory works has a post of its own.
- Opening a screen for you. It changes no data.
- The seasonal effect. Falling snow, blossom petals or tumbling leaves on this device. Ask her to switch it off.
Ask for the same change on several things at once and it comes back as one thing to approve. "Archive these three." is read back as "Archive 3 threads", and one yes covers all three. On a call, when a change deletes something, the button says Delete, in red.
It was built this way so you can talk loosely and still be in charge. You don't need careful commands, because every change that matters is read back before it happens. And a batch is one decision, not three interruptions.
What happens when you say no?
Nothing changes. Say no, or tap Cancel, and Ranya drops the change. Your tasks, notes and inbox stay exactly as they were.
Example call. You say: Archive these three. Three emails sit in your inbox: the weekly digest, your gym and the accountant. Ranya is asking: Archive 3 threads, with Cancel and Confirm. Cancel leaves all three in the inbox; Confirm archives them.
Then say what you meant. "No, keep the one from the accountant." is a perfectly good answer. Whatever she proposes next is read back the same way, so the corrected version waits for your yes too. A no is never a dead end.
How to stay in charge of an AI assistant, step by step
Five habits, from the first question of a call to the day after.
Ask anything first
Anything Ranya only reads runs on its own: "What's on today?", "What's slipping?", "Any new mail?". She looks it up and tells you. Nothing changes, so nothing is asked. If you have never called her, start with how a call with Ranya works.
Listen, then answer
When you ask for a change, she reads it back, and the card on screen says Ranya is asking. On a computer it also says Or just say "yes". Say yes or tap Confirm. Say no or tap Cancel.
Group the small stuff
"Complete these three." or "Archive these." comes back as one question, not three. Clear a list with one yes, and keep your attention for the changes that deserve it.
Read the transcript afterwards
Every call is saved to your chats as a written transcript of both sides. Open it to see what was asked, what you agreed to and what you turned down.
You: Any new mail?
Ranya: Three: the weekly digest, your gym and the accountant.
You: Archive these three.
Ranya: Archive all three?
You: No, keep the one from the accountant.
Archive threadDeclinedRanya: Just the digest and the gym, then?
You: Yes.
Archive thread2 threadsExample transcriptThe call as it reads in your chats afterwards: both sides, the change you declined and the one you approved. Deleted something you needed? Open the Trash
Deleted tasks, notes, goals, habits and chats all go to one Trash, and they stay there until you delete them forever. Open it and tap Restore, or ask on a call: "Restore the proposal task." That is a change too, so she reads it back first.
Example TrashOne Trash for everything, kept until you delete it. Example Trash with three things deleted two days ago: the task Finish the proposal, the note Trip ideas and the goal Read 2 books. Each shows when it was trashed and a Restore button.
Can you trust an AI assistant with your personal information?
Asking first covers what an assistant does. The other half of trust is what happens to what you tell it. These are Ranya's promises, and each one is in the privacy policy.
Every call is saved to your chats as a written transcript. Your voice is never stored as a recording.
We don't sell your information and don't use it for advertising.
We don't store your mailbox. Messages are fetched from Gmail when you open them.
Your Gmail data is never used to train AI.
Delete your account yourself, any time, from Settings.
Our privacy policy names every service we use and what each one receives.
To answer you, Ranya sends what you asked, and the items it needs, to the AI providers it uses. The privacy policy names each one and what it receives. See everything Ranya promises about your data.
Asking first covers email too: drafts never send themselves, and an email goes out only when you send it or say yes to sending it.
Frequently asked questions
Is it safe to let an AI assistant change my tasks, calendar or email?
It is safest when the assistant has to ask before it changes anything. In Ranya, every change to your tasks, habits, goals, notes or email is read back first and waits for your yes or a tap on Confirm, and saying no means nothing happens.
What does "human in the loop" mean?
It means a person approves an important action before an automated system carries it out. For an assistant, that means it can look things up on its own, but it asks you before it creates, changes, sends or deletes anything that matters.
What can Ranya do without asking?
Look things up, remember something you told her, open a screen for you, and switch a seasonal visual effect on your device. Everything else waits for your yes, adding a task to a focus included, since she asks which focus first.
What happens if I say no?
Nothing changes. Ranya drops the change, and you can tell her what you meant instead.
What if I approve something by mistake?
Deleted tasks, notes, goals, habits and chats go to one Trash, and stay there until you delete them forever. A sent email cannot be unsent, so sending is always its own question. Every call is also saved to your chats as a written transcript, so you can see exactly what was agreed.
Does Ranya sell my data or use it to train AI?
We don't sell your information or use it for advertising, and your Gmail data is never used to train AI. What you ask about is sent to the AI providers named in the privacy policy, which handle it under their own terms.
Keep reading
- Talk to Ranya
A Voice Planner You Can Call: Plan Your Day Out Loud
A lot of planning happens on a walk or at the sink, not at a keyboard. How to plan your day out loud, with a planner that reads each change back and waits for your yes.
· 9 min read
- Talk to Ranya
An AI Assistant That Remembers, With Notes You Can Edit
An assistant should remember what matters and show you all of it. How Ranya's memory works, where it shows up, and how to edit or delete any note.
· 8 min read
- Talk to Ranya
An AI Assistant for Your To-Do List That Keeps the Plan
A chat can write a lovely plan that is gone by Wednesday. What changes when the assistant can put the plan where you act, and keep it there.
· 10 min read