Encrypted in transit and at rest. Never sold, and never used to train AI models. Your AI employee asks before it acts as you, and keeps a record of everything it does.
Independently assessedCASA Tier 2
Stored inthe United States
Never used totrain AI models
What we promise
Four promises, kept in the product.
Control
You approve before it acts as you.
From your own inbox, replies wait for your approval by default
You choose whether outreach from its own address waits for you too
Errands on its computer wait for your go-ahead, and you can watch or take over
Purchases, deposits and card holds always wait for you
Visibility
You can see everything it does.
A record of every reply drafted, sent or held, and why
From its own address and on calls, it says it’s an AI
A morning brief of what it handled overnight
Data protection
Your data is protected.
Encrypted in transit and at rest
Bank, medical and legal mail in your inbox is skipped before AI reads it
Carries out the outreach you ask for, from its own address, copying you
Prepares your morning brief
Answers calls from people it knows, and says it’s an AI
Asks you first
Replies from your own inbox, by default
Outreach too, if you switch it to approval
Any errand on its computer, before it starts
Every call it places, and anything that commits money
Never
Reads bank, medical or legal mail in your inbox with AI
Sees your passwords: you sign in to sites yourself
Shows your data to another account
Trains AI models on your data
Pretends to be a person
These are the defaults. You can require approval for more, channel by channel, in Settings.
Follow one email
Sensitive mail in your inbox stops before AI.
Every email in your connected inbox is checked before the AI that reads and drafts sees it. Mail that looks financial, medical or legal stops there. By default, a draft goes out from your inbox only after you approve it. Mail sent to your employee’s own address, and a business inbox a specialist runs, are read for the work and aren’t checked this way.
Independent assessments
CASA Tier 2
Independently assessed
CASA Tier 2
The App Defense Alliance’s security assessment for apps that access Google user data. Atona’s web app was assessed in April 2026. It is an independent assessment, not a Google certification.
Tell us what you found and the steps to reproduce it.
We reply to every report and keep you updated until it’s fixed.
We won’t pursue good-faith research that avoids other people’s data and gives us time to fix the issue.
Running a security review? Ask the same address for our documents and a completed questionnaire.
Common questions
Does Atona read all my email?
It sorts incoming mail in your connected inbox so it can draft replies and brief you. Mail from banks, tax offices, insurers, healthcare providers and law firms, or with subjects like statements, tax forms and one-time codes, is recognised by its sender and subject and skipped before the AI that reads and drafts sees it. When a message’s body looks like it holds personal identifiers, a short excerpt is checked by a classification model first.
Is mail to my employee’s own address checked the same way?
No. Mail people send to your employee’s own address is the conversation it is handling, so it reads it. The same goes for a business inbox a specialist runs, like a hosting or brand-deals inbox: that mail is the job. Connect an inbox you use for that work, not your personal one.
Can people tell they’re dealing with an AI?
Yes, when it writes from its own address: its emails are signed as your AI assistant. On calls it says it’s an AI and never denies it. Replies you approve from your own inbox go out as you.
How long do you keep my data?
What it remembers from conversations is deleted after 90 days. The rest stays while your account is open and goes when you delete it.
Is Atona built for health information?
No. Atona isn’t a HIPAA service, and medical mail is held back from AI.
Is Atona available in Europe?
Not yet. Atona isn’t offered in the EU, EEA, UK or Switzerland.