Can your keyboard see your AI chats? What Gboard, SwiftKey, Samsung, and Grammarly send
The Morse teamPublished 10 min read
Yes. A keyboard app sees every character you type before the chat app does, AI chats included, and Android's own warning says a keyboard "may be able to collect all the text you type" (Android source). As of November 2026, Google (Gboard Help) and Microsoft (Microsoft) say Gboard and SwiftKey keep ordinary typing on your phone by default, but Gboard, SwiftKey, and Samsung Keyboard all have AI writing features, sync, or data-sharing settings that send text off the phone for cloud processing, and Grammarly's keyboard uses Grammarly's servers to check your text at all (Grammarly).
We read what each company says on its own pages on November 16, 2026: Gboard's help center and App Store listing, Microsoft's SwiftKey help pages and its November 2026 privacy statement, Samsung's privacy policy and Galaxy AI pages, Grammarly's help center, and Apple's and Google's documentation for keyboard developers. Some defaults earn credit: Apple cuts third-party keyboards off from the network unless you allow it, Android warns you before you turn one on, and Microsoft's default for SwiftKey is that nothing it learns about your typing leaves the device.
Can a keyboard app see everything you type?
Yes, in any field it's allowed to type into. When you switch on a new keyboard on Android, the system shows this warning: "This input method may be able to collect all the text you type, including personal data like passwords and credit card numbers" (Android source).
Seeing your text and sending it are two different things. On an iPhone, Apple's Platform Security guide says custom keyboards "run by default in a very restrictive sandbox that blocks access to the network," so they can see your typing but not send it anywhere. The same page says a keyboard is used "for any text field except the passcode input and any secure text view," where the system swaps in Apple's own keyboard, and that app developers "can prevent the use of third-party keyboard extensions within their app." Apple's guide for keyboard developers gives banking apps as an example of apps that do this (Apple).
What does "Allow Full Access" mean on an iPhone?
It lets a third-party keyboard reach the internet. Apple's developer guide sets out the trade in a table: with open access off, "users know that keystrokes go only to the app that is using the keyboard," and with it on, the "keyboard can send keystrokes and other input events for server-side processing" (Apple). Without it, Apple says, "the system ensures that keystrokes cannot be sent back to you or anywhere else."
Apple's App Review Guidelines, last updated October 16, 2026, require keyboards to "remain functional without full network access and without requiring full access," and to "collect user activity only to enhance the functionality of the user's keyboard extension on the iOS device." Rule 2.5.18 keeps display ads out of keyboards too.
Grammarly shows why some keyboards ask for it anyway. Its help center says the Grammarly Keyboard "needs full access in order to connect to the internet" because "we use powerful servers to check your text," and that without it you can type but get no suggestions (Grammarly). To check, open Settings, tap General, then Keyboard, then Keyboards, tap the keyboard, and look at Allow Full Access.
Does Gboard send what you type to Google?
On an iPhone, Google says it doesn't, apart from searches and voice. Gboard's App Store listing says "Other than your searches and voice inputs, Gboard doesn't send anything you type to Google," and adds that Gboard sends usage statistics and that microphone input "will be sent to Google for transcription" (App Store).
That line describes Gboard on the iPhone. On Android, Gboard has AI features that can use Google's cloud:
- Writing tools (Proofread, Rephrase, Shorten, and others) run "in a secure, isolated environment on your device or in the cloud," and Google says "your input text isn't shared with Google or any app" (Gboard Help).
- Personalized suggestions on the Pixel 11 series go further. With it on, Gboard "can use your typing history, as well as screen context taken at the time of use from supported first and third-party messaging apps," processed "through secure cloud servers with Private AI Compute" (Gboard Help). Google names Google Messages, Google Chat, and LINE, and says it isn't offered in the European Economic Area. We cover what Google's enclave system promises in Google Private AI Compute.
- Advanced voice typing on Pixel 6 and later keeps what you say on the phone "except when you use the 'Fix it' or detailed edits features" (Gboard Help).
Gboard also learns from your typing through federated learning, which is on by default and sends Google "what it learns" rather than your text (Gboard Help). What is federated learning explains how that works and how to switch it off. Gboard's Delete learned words and data button, under Privacy in its settings, "clears all on-device data that Gboard has saved."
Is SwiftKey safe?
By default, SwiftKey keeps your typing on the phone, and three choices change that. Microsoft's help page says "by default, nothing Microsoft SwiftKey learns about your typing leaves your device" (Microsoft). The three choices:
- Help Microsoft to improve. If you agree when SwiftKey asks, it captures "snippets" of your typing, anonymously and without sign-in. Microsoft's privacy statement, last updated November 2026, calls them "short, de-identified snippets of what and how you type" that are "never linked to your SwiftKey account" (Microsoft).
- Signing in. With a Microsoft or Google account, SwiftKey "will then periodically send fragments of data about the words you use to a Microsoft cloud service" (Microsoft). The November 2026 privacy statement says backup and sync now saves your language model "to OneDrive," and needs a Microsoft account.
- Copilot. SwiftKey's Tone feature rewrites your text as Professional, Casual, Polite, or Social Post, and Microsoft says "to use the Tone and Chat features, you must sign in with your Microsoft account" (Microsoft). The page doesn't say where the rewriting happens.
Microsoft is candid about password fields. SwiftKey "does not learn anything from fields marked as password fields," but "if you use the 'show password' option on an app or website, that text field will no longer be identified as a password field" (Microsoft). It also warns that a password you once typed into a text or email can be learned. Deleting your SwiftKey account data "will not affect your dynamic language model stored on your device," so you clear that separately on the phone: Microsoft's iOS steps include Remove my data from this device (Microsoft).
Does Samsung Keyboard send your text to Samsung?
Its AI features do. Samsung's Writing assist lives in the Samsung Keyboard on recent Galaxy phones and rewrites emails, texts, and posts. Samsung's account privacy notice, effective February 24, 2026, lists Writing assist's "entered or selected text" as input processed outside your device (Samsung).
Ordinary typing is a separate entry. Samsung's privacy policy, effective December 31, 2025, lists "Samsung Keyboard information: The words that you type when you enable 'Predictive text,'" which "may be offered in connection with your Samsung account to synchronize the data for use on your other Samsung mobile devices" (Samsung). You can clear it in the Predictive text settings.
To keep Galaxy AI on the phone, open Settings, tap Galaxy AI, and turn on Process data only on device, which Samsung calls a "master switch" for cloud processing (Samsung). Is Galaxy AI private covers what that switch misses.
Keyboard app privacy, side by side
What each company's own pages say, read November 16, 2026.
| Keyboard | What leaves the phone by default | AI features that send text | Training or learning | Switch to turn it off | Source |
|---|---|---|---|---|---|
| Gboard on Android | What federated learning learns, not the text | Writing tools, "on your device or in the cloud"; Pixel 11 Personalized suggestions, with screen context | Federated learning on by default; audio donations opt-in (Google) | Gboard Settings, Privacy; Writing tools menu, Personalized suggestions | |
| Gboard on iPhone | Searches, voice input, and usage statistics | None named in the listing | Not stated | Allow Full Access off | App Store |
| Microsoft SwiftKey | Anonymous usage stats (Microsoft); nothing it learns about your typing | Copilot Tone and Chat, with a Microsoft account | Typing snippets, only if you opt in | Privacy, Help Microsoft to improve; turn off Backup & Sync | Microsoft |
| Samsung Keyboard | Not stated; Predictive text words may sync with your Samsung account | Writing assist, with "entered or selected text" | Not stated in the notices we read | Galaxy AI, Process data only on device | Samsung |
| Grammarly Keyboard | The text you type with it, to Grammarly's servers | Every suggestion | On by default for Free, Premium, and single-user Pro | Product Improvement and Training toggle | Grammarly |
| Apple keyboard on iPhone | Dictation audio, unless your settings show it's processed on the device | Not covered here | Usage reports only if you opt in, with differential privacy | General, Keyboard, Enable Dictation | Apple |
Apple's Siri and Dictation page says your device "will indicate in Siri Settings or Keyboard Settings whether the things you say are processed on your device and not sent to Apple," and that otherwise "the things you say are sent to and processed by Apple" (Apple). Apple's 2017 research note on keyboard and emoji statistics says its system is "designed to be opt-in" and randomizes data on the phone "so the server never sees or receives raw data" (Apple Machine Learning Research). Apple Intelligence's writing features are covered in is Apple Intelligence private.
Does incognito mode stop your keyboard from learning?
Only if the keyboard honors it. Android lets an app ask the keyboard not to learn from a field, with a flag Google describes as a request that the keyboard "should not update any personalized data such as typing history" (Android Developers). Google's documentation adds: "Applications need to be aware that the flag is not a guarantee, and some IMEs may not respect it."
The flag covers learning only. A keyboard in incognito still sees what you type, and a cloud AI feature you tap still sends the text you asked it to rewrite.
Does end-to-end encryption protect what your keyboard sees?
No. A chat app's encryption starts once the text reaches the app, and the keyboard handles it before that. If a keyboard sends your words to its own servers, the chat app's encryption never touches that copy.
What a keyboard sends can also leak on the way. On April 23, 2024, the Citizen Lab reported that cloud-based pinyin keyboard apps from eight of the nine vendors it tested, Samsung among them, had flaws that could "completely reveal the contents of users' keystrokes in transit" (Citizen Lab). Most vendors fixed them, and the researchers advised at-risk users to switch to a keyboard "that operates entirely on-device."
How do you check your keyboard before a sensitive chat?
Work through these before you type:
- Find out which keyboard you're using. On an iPhone, look under Settings, General, Keyboard, Keyboards. On Android, Google's path is Settings, System, Language & input, then On-screen keyboard (Gboard Help).
- Turn off Full Access on any iPhone keyboard that doesn't need it, or switch to Apple's keyboard for sensitive typing.
- Turn off the cloud AI features you don't use: Gboard's Personalized suggestions, SwiftKey's sign-in and Backup & Sync, Samsung's Writing assist, or Grammarly entirely.
- Check the learning switches: Gboard's Privacy screen, SwiftKey's Help Microsoft to improve, and Samsung's Predictive text data.
- Type rather than dictate anything sensitive, unless your phone says dictation runs on the device.
Before you type anything sensitive into any chat, use a keyboard that keeps text on the device. The same habits apply to browser add-ons that read the page you're typing on, which we cover in are AI browser extensions safe.
How Morse keeps your chats private after you send them
Once you press send, the questions move to the chat service: who reads the message, who keeps it, and who could be asked to hand it over. Morse answers them with a privacy mode you pick for each chat.
- E2EE, for anything you'd show no company. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. It runs open-weight models like GLM, Kimi, and DeepSeek, rests on the enclave hardware and attestation working as designed, and has no web search or image generation, since models can use only tools that run on the device. The enclaves are run by third-party providers, not by Morse. E2EE chats come with Pro, at $19 a month.
- Private, for everyday drafts and questions. It's the default for a new chat. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing, and where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
- Anonymous, for GPT, Claude, or Gemini without your account. Morse's proxy removes your account, email, and IP address, so the provider sees the message without knowing it's you. It may keep the message for a limited time to check for abuse, under its own policy, so leave out names you'd rather keep to yourself.
In every mode, your chat history is stored in your browser, encrypted, not on Morse's servers. Morse doesn't train AI on your chats, sell your data, or show ads, and because we don't store chats, we usually have very little to give authorities (Morse privacy). Clearing your browser deletes your chats, and Morse can't recover them, so keep a copy of anything you need. For the rest of the privacy checklist, see how to use AI privately.
Try a chat from the Morse home page without an account, or start free with no card needed.
Common questions
Can a keyboard app read my passwords?
On an iPhone, no third-party keyboard is used in the passcode field or in any text field marked secure, so Apple's keyboard takes over there. On Android, the system warns when you turn on a keyboard that it may collect passwords and credit card numbers. SwiftKey says it doesn't learn from password fields, but a field stops counting as one once you tap "show password."
Does deleting my keyboard's account delete what it learned?
Not always. Microsoft says deleting your SwiftKey account and data doesn't touch the language model stored on your phone, which you clear separately. Gboard's Delete learned words and data button clears what Gboard has saved on the device.
Which keyboard is the most private?
One that can't reach the network. Apple says a third-party keyboard without Full Access can't send your keystrokes anywhere. On Android, check a keyboard's privacy settings and turn off its cloud AI features before a sensitive chat.
Can Gboard read the messages on my screen?
On a Pixel 11 with Personalized suggestions turned on, yes. Google says Gboard's Personalized suggestions then use screen context from supported messaging apps, such as Google Messages, Google Chat, and LINE, and process the request in the cloud through Private AI Compute. Turning Personalized suggestions off stops it.