Should an AI ask before using a saved password?
Yes. An agent that types a saved password should ask first, by default, with one tap. Sites lock accounts after repeated failures, so it should try that password only once. GROX asks “use my saved login?” before its computer signs in, tries a saved password only once, and you can turn the asking off in chat.
Why should an agent ask before typing a saved password?
A saved login is convenient, but it is still a secret that belongs to you. If an agent types it the moment a form appears, you lose the chance to notice the wrong site, a look-alike page, or a session you did not mean to continue. One tap is a pause, not a lecture.
The model should never see the value. The password belongs on the page it is for, not in chat history. Asking first also keeps the human in the loop when the agent is about to act on a real account. A simpler password manager in the browser is still the better choice if you only need fill-in on a machine you already control, with no cloud computer in between.
What goes wrong if it retries a refused password?
Many sites treat a second or third wrong password as an attack. They lock the account, force a reset, or flag the login. An agent that keeps trying “just in case” can lock you out of a service you still need. One try, then stop and ask, is the safer pattern.
Turning the asking off is a choice, not a requirement. Some people want silent sign-in on a machine they already trust. Others want every login named out loud. The default should be the ask, because the cost of one extra tap is small compared with a lockout or a password typed into the wrong place.
| Approach | Ask first | Tries if refused | Who sees the secret |
|---|---|---|---|
| Browser password manager | Usually not, after you save it | Browser policy; often more than one | The browser, not a chat model |
| Agent with no pause | No | Often retries until it succeeds or fails | Risk of the value landing in logs |
| GROX | Asks “use my saved login?” with one tap; asking can be switched off in chat | Tries a saved password only once, then stops and asks you | Password goes into the page, never into the chat and never to the AI |
How does GROX ask before using a saved login?
When its computer reaches a sign-in that already has a saved login, it asks “use my saved login?” with one tap before it signs in. You can turn that asking off in chat if you want it to proceed without the card. It tries a saved password only once. If the site refuses it, it stops and asks you rather than guessing again.
Passwords never reach the model. When a site on that computer asks you to sign in, the chat shows a secure form. The password goes straight into that page, never into the chat and never to the AI, and is typed only on the site it is for. Tick “save it for next time” and it is kept encrypted so sign-in can happen by itself later; otherwise it is forgotten within 30 minutes. An API key goes in the same way and reaches the terminal without being shown.
The computer works one visible step at a time. When it reaches a password, a one-time code or a payment page, it can stop and send your phone a link that opens only that screen for fifteen minutes. Done hands the work back. The link is not a sign-in to your account.
- One tap
- Before the computer signs in with a login you saved, it asks “use my saved login?” unless you turned the asking off in chat.
- One try
- It tries a saved password only once. If the site refuses it, it stops and asks you.
- Never in chat
- The password is typed only on the site it is for; it never goes into the chat and never to the AI.
- Optional save
- Tick save for next time and it is kept encrypted; otherwise it is forgotten within 30 minutes.
When is a simpler tool the better choice?
If you only fill forms on your own laptop and never want a cloud computer to browse for you, the browser's own manager is enough. You do not need an agent in that loop. Use the cloud computer when the work is browsing, signing in one step at a time, or handing a password page to your phone.
Chat actions that send, post, pay or change something still wait for Approve. Reading and drafting can run without a card. A saved login is a middle case: it is not a payment, but it is still a secret, so the default ask is the right trade-off.
Common questions
Should an AI agent ask before it uses a saved password?
Yes. Ask first by default, with one tap, because the site might be wrong or the session unintended. Try the saved password only once so a refusal does not lock the account. The model should never see the value. Turning the asking off can be your choice after you understand the risk.
How does GROX handle a saved login on its computer?
Before the computer signs in with a login you saved, it asks “use my saved login?” with one tap. You can turn the asking off in chat. It tries a saved password only once; if the site refuses it, it stops and asks you. The password never enters the chat or the AI.
Does the AI see the password I type?
No. A secure form in the chat sends the password straight into the page. It is never in the chat and never shown to the AI. Save it encrypted for next time, or it is forgotten within 30 minutes. An API key uses the same path into the terminal without being shown.
Why only one try if the password is refused?
Sites often lock accounts after repeated failures. An agent that retries a saved password can lock you out. GROX tries a saved password only once, then stops and asks you. That matches how people should treat a refused login: stop, check the site, then type again yourself if needed.
Read how sign-in and saved logins work in the Help Centre or start from Home.