"Is it safe?" is the question everyone asks about AI keyboards, and the one few articles answer, because safety here is not one property. It is a handful of separate things you can check. Some of them iOS guarantees whichever keyboard you pick. Others depend on how one developer built one app, and you have to look them up.
Here is the checklist, split the same way.
Two risks, not one
What the keyboard sees. Every keyboard sees what you type into it, because that is how a keyboard works. The question is whether that text stays on your phone or goes somewhere, and for what.
What the AI feature sends. A keyboard that drafts a reply or rewrites a sentence has to hand that text to a model. Unless the model runs on the phone itself, the text travels to a server. So for an AI feature the useful question is not whether anything is sent, which the feature needs, but what is sent, when, and whether you control it.
Mixing the two up leads people to reject every AI keyboard, or to trust one because it feels private. Check them separately.
What iOS already guarantees
This part does not depend on the app. Apple's App Extension Programming Guide says that when you tap a secure field, the kind that shows characters as dots, iOS temporarily replaces the custom keyboard with its own. The same happens in phone number pads, and an app can refuse custom keyboards altogether. That applies to every third-party keyboard, with Full Access on or off.
Apple's page on configuring open access for a custom keyboard adds the other half. Without Full Access, a keyboard has no network access, so your keystrokes go only to the app you are typing in. A keyboard can send text to a server only after you turn on Allow Full Access for it in Settings.
So the password question is handled by the platform. What is left to check is everyday typing: messages, emails and notes.
What you have to verify, app by app
None of this is guaranteed by iOS. A good developer says it plainly.
Is the AI opt-in? A keyboard where AI processing has to be switched on before anything is sent is different from one that is live the moment you install it. Look in the app's own privacy settings, not just the store description.
What triggers a request? Ideally, only the moment you tap a feature. If the privacy policy cannot describe a trigger, note that.
What is sent? Only the text you are working on, or a wider slice such as conversation history and contact names? The narrower the scope, the less there is to go wrong.
Are sensitive details filtered? Some keyboards strip phone numbers, email addresses, links and one-time codes out of text bound for an AI provider. That is a specific claim you can check. Either the developer states it or they do not.
Who processes it? A good consent screen names the companies that will handle your text, not just "our partners".
Does it need an account? A keyboard you can use without a login leaves a smaller trail tied to your identity. An account can also make it easier to see and delete your data, so weigh both.
Can you clear what is saved? If the keyboard remembers things about your contacts to improve suggestions, check that you can see what is stored and delete it.
A five minute audit for the keyboards you already have
Open Settings, then General, Keyboard and Keyboards. For each keyboard from another developer:
- Check whether Allow Full Access is on. If it is off, that keyboard cannot reach the network.
- Open the keyboard's own app and look for a privacy section with a separate switch for AI.
- Search its privacy policy for "opt-in", "off by default" and "redact". Reassurance without a mechanism is a reason to keep reading.
- Look for the names of the companies that process your text.
- Note whether you needed an account for basic features.
It takes about as long as reading a page of reviews and tells you more. For what the switch itself does, see What Allow Full Access means for an iPhone keyboard.
How Protexter answers the checklist
Here is how the keyboard this site is about answers each item, so you have a real example to compare with:
- Opt-in AI. Third-party AI processing is a switch in the app, off by default, and the keyboard needs Full Access as well. Conversation context and automatic contact memory are separate switches, both off by default.
- Trigger. Requests go out when you tap a feature such as Fix, Rewrite or Ask AI. Automatic reply suggestions run only if you have also turned on conversation context.
- Who processes it. The consent screen names Vercel, OpenRouter and the model provider. Every request carries a zero data retention requirement, which limits it to model endpoints that support that policy.
- Filtering. By default, phone numbers, email addresses, links and one-time codes are redacted from conversation context. You can turn that off when you need the exact details.
- Screenshots. Text in a screenshot is read on your phone, and the image is not uploaded.
- Account. There is no account or login to create, and no API key to set up. Setup asks for a name and email address for setup tips, and you can skip that step.
- Clearing. The app's Privacy screen has one action that clears local app data, including saved voices and saved keyboard drafts.
Do not take this on faith either. It is exactly what the checklist tells you to verify, for this keyboard and any other.
The short version
Safe is not a single yes or no. iOS keeps every third-party keyboard out of passwords and secure fields, whatever you install. What is left is checkable: is the AI opt-in, what is sent when you use it, who processes it, are sensitive details filtered, and can you clear what is saved. Run the checklist before you decide, rather than going by a star rating or a feeling.