Secure AI has no password field. Not a hidden one, not a fallback for people who ask nicely — there is no code in the product that could accept a password if you typed one. Sign in goes through Apple, Google or Microsoft. This is what that buys, and what it costs.
The database we do not have
Every custom login system is, underneath, a table of people and their secrets. Even done properly — the secrets hashed with a slow algorithm, the salts unique, the work factor current — that table exists, it is worth money to somebody, and it is the first thing anyone looks for. Most breaches you have read about are that table.
The strongest thing you can say about a store of credentials is that it is well defended. The stronger thing is that there is nothing in it. Ours is empty because it was never created.
A password we never receive is a password that cannot leak from us.
What goes away with it
The password is only the visible part. Building your own login means building everything that hangs off it, and each piece is its own way in:
- •The reset flow. “Forgot your password” is a mechanism for changing somebody’s credential using only their email address. It is the route attackers take most often, because it is the one part of a login system designed to let a stranger in.
- •The reset email. A message with a link that grants access, arriving in an inbox, teaching people that such messages are normal. That lesson is what makes phishing work.
- •Security questions. Facts about you that are easier to find than to remember.
- •Credential stuffing. Passwords reused across sites get tried in bulk against any login form. Ours does not exist, so there is nothing to try them against.
None of these are hypothetical failures of careless companies. They are the standing cost of running a login form, paid by everyone who runs one.
What you get instead
Handing sign-in to Apple, Google or Microsoft is not only avoiding work. It attaches your Secure AI account to security machinery none of us could build alone, and all of it runs before we are told who you are:
- •Two-factor authentication, already on. If your provider account uses a code, a prompt on your phone, a passkey or a physical security key, that check happens at sign-in. There is no separate Secure AI second factor to enable, and — more to the point — no Secure AI login that could bypass the one you already have.
- •Location and device risk. Providers score every sign-in against where and how you normally sign in. A login from a new country, a new device, or an implausible jump between the two gets challenged. Reproducing that would mean us tracking your location, which is the sort of thing this product exists to avoid.
- •Breach monitoring. They check your credential against known breach corpora and force a change when it appears. We would have no way of knowing.
- •One place to revoke. Your provider lists every device and app with access, and can cut any of them off. That list includes Secure AI, and revoking there ends it here.
- •Recovery run by people who can do it. Losing access to an account is where identity gets proven or faked. Apple, Google and Microsoft have staffed, adversarial processes for this. Our version would be somebody reading an email and making a judgement.
How it works, technically
The mechanism is OpenID Connect, the identity layer on top of OAuth 2.0. Sign-in sends you to the provider on their own domain — you are typing into Apple’s page, or Google’s, or Microsoft’s, never into ours. They authenticate you however your account is configured, and hand back a signed identity token.
That token is a short-lived assertion, cryptographically signed with the provider’s private key, saying which account signed in and when. It is verified against the provider’s published keys. The signature is what makes it useless to steal in transit and impossible to forge: nothing in it can be altered without breaking it.
What we keep from the exchange is an account identifier, plus the display name and email address the provider chooses to share. Every later request carries a short-lived token that our servers verify on arrival, so access can be withdrawn without waiting for a session to expire. At no point does a password exist in a form we could store, log, or lose.
Sign in with Apple, and the address you never gave us
Apple goes one step further with Hide My Email. Choose it and Apple generates a unique relay address for Secure AI, forwards mail to your real inbox, and never tells us what that inbox is. You can switch the relay off later and the mail stops, without touching your real address and without asking us for anything.
It is worth being plain about what that means for us: for those accounts we cannot email you outside the relay, cannot match you to a marketing list, and cannot recognise you from an address seen anywhere else. That is the correct arrangement, and it is available because the provider offers it, not because we promised to behave.
The honest cost
Three things are worse under this design, and you should weigh them.
Your Secure AI account depends on an account somewhere else. Lose the provider account and you lose the route in. This is the main reason three providers are offered rather than one — most people have at least two, and the two are not held by the same company.
The provider knows you use Secure AI. They see the sign-in, though not the conversation: what is asked, what is answered, and what is stored never touches them. Set against what a password database gives away when it leaks, this is the trade we would make every time.
And somebody with your unlocked phone and an already-signed-in provider account is signed in here too. That is why the phone’s own lock matters, and why Secure AI can be locked separately on the devices that support it.
Or don’t sign in at all
Guest mode needs no account of any kind. Nothing syncs between devices and the session ends when you leave it, which is the trade, stated rather than buried. Signing in later keeps the conversations you have already had.
Questions
Will you add email and password later?
No. Adding it would recreate every problem above, and it would do so for the accounts already here — a login system is only as strong as its weakest route in, so a password option would become the way into accounts that never used one.
Is my provider password stored anywhere in Secure AI?
It is never sent to us. You type it on your provider’s own domain, and what we receive is a signed token saying they are satisfied it was you.
What if I signed up with Google and want to use Apple?
Sign in with the other provider using the same email address and it resolves to the same account. Where the addresses differ — an Apple relay address, for instance — they are separate accounts, because nothing links them that we could verify.
Does this mean Google or Apple can read my chats?
No. They confirm who you are and nothing else. Conversations are not sent to them, and the account identifier they give us reveals nothing about what is in one.
Try Secure AI free
Frontier AI chat — encrypted by default, with your identity stripped before any request reaches the provider.