Is Grammarly safe? What it can see, what it keeps, and how to switch it off

The Morse teamPublished November 9, 202610 min read

As of November 9, 2026, Grammarly is safe in the security sense: it's a long-running product with a SOC 2 Type 2 audit, five ISO certifications, and encryption in transit and at rest. It isn't private by default. Grammarly reads the text in every field where it's active, and it trains on that text unless you switch off "Product Improvement and Training" (Grammarly). Its parent company, Superhuman, keeps most of what it collects for periods its policy doesn't name (Superhuman). It isn't a keylogger by its own definition, but its opt-in Authorship feature records "your full typing and editing history," deleted text included, and keeps it for up to 10 years once you open a report (Grammarly).

We read Grammarly's privacy and trust pages, its security page, Superhuman's privacy policy and sub-processor list, Grammarly's terms, about 15 articles in its help center, and Google Project Zero's 2018 report on November 9, 2026. Grammarly deserves credit for answering hard questions in plain words: its help center publishes a "Does Grammarly's product read everything I write?" page, names the auditor behind its SOC 2 report, and points every user to a sub-processor list that names its AI providers.

Grammarly's safety and privacy at a glance

Read November 9, 2026.

Is Grammarly a keylogger?

No, by Grammarly's definition, which is a tool that records every keystroke "without the user's knowledge" and sends it to a third party. Grammarly's help center says "Grammarly does not record every keystroke on your device" (Grammarly). The same page says Grammarly is "blocked from running in read-only and sensitive fields," and that "you can easily turn it off any time."

Authorship is the exception you opt into. It tracks your writing in Google Docs through the browser extension, and in Microsoft Word through the desktop app (Grammarly). It's off by default in Docs and Word, and a colored thumbprint icon in the lower-left corner shows when it's on. In docs, Grammarly's newer AI writing surface, Authorship is "enabled automatically, capturing every draft and revision from the start."

What Authorship collects goes past the finished text. Grammarly says it keeps "text you write and then delete," text you paste along with the names of the sites it came from, such as Wikipedia and ChatGPT, and "your prompts to generative AI" (Grammarly). With clipboard access, it sees the latest text you copied in your browser, or in any desktop app when you work in Word. Its report includes a replay of your typing that you can leave out before sharing. Grammarly also says that Authorship data on the device "may be accessible by other users of the device while you are logged in," so sign out on a school or library computer.

Does Grammarly read everything you type?

It reads what you type wherever it's switched on, and skips sensitive fields such as passwords and payment details only on a best-efforts basis, as Grammarly alternatives sets out from Grammarly's privacy page. On a phone, Grammarly says a gray icon means you're typing in a sensitive field, such as a password, email, or browser search field (Grammarly). Don't rely on it for an account number typed into an ordinary text box.

You choose where it runs, at three levels:

• One text field: hover over the Grammarly button in the field's lower-right corner, click the power button, and choose to turn it off until your next visit or forever (Grammarly).
• One website: on that site, click the Grammarly icon in the browser toolbar and turn off Allow extension to check text on… (Grammarly). To stop it everywhere, open a new tab and turn off Check text with browser extension.
• One desktop app: in Grammarly for Windows, right-click the floating widget; on a Mac, click the menu bar icon. Choose Turn off Grammarly, and manage the list under Settings > Block list (Grammarly).

How any extension can read the pages you visit is in are AI browser extensions safe.

Is Grammarly safe for confidential information?

Not on its defaults: on Free, Premium, and single-user Pro accounts, Product Improvement and Training is on until you switch it off (Grammarly), and Grammarly alternatives gives the steps. Grammarly's help center adds that it may use "random samples of aggregated and de-identified user content" unless you've opted out (Grammarly).

Business accounts aren't all excluded. Training starts off for Grammarly for Education, Business, Pro, and Enterprise accounts "purchased through our Sales team" (Grammarly). A multi-user team account bought on Grammarly's website starts with training on, and only its admin can switch it off. If you leave a team for a Free account, the team's setting carries over to your account, so check the toggle again.

A second setting keeps a profile of your writing. With "tailored assistance" on, Grammarly will "process and store data and associate it with your account" to personalize suggestions (Grammarly). Turning it off at account.grammarly.com/customize/features deletes the data stored for that feature, Grammarly says.

Health information needs a contract that most plans can't get. Grammarly says that under its Acceptable Use Policy you shouldn't process protected health information unless you've signed a Business Associate Agreement with it (Grammarly). It signs those agreements "only for Grammarly for Business Enterprise plans." This isn't legal advice, so ask your compliance team before you check a patient letter.

Who else can see your Grammarly text?

Four AI companies for its generative features, and its own staff in narrow cases. Grammarly's help center says that prompt text and "the context in which it's used" go to "our small number of thoroughly vetted service providers," which it doesn't let train on user content (Grammarly). Superhuman's sub-processor page names them as Anthropic, Azure, Baseten, and OpenAI, all in the USA (Superhuman). That page describes processing for business customers, and the help center points every user to it (Grammarly). Grammarly says its generative AI features "are not turned on by default," and you can turn them off under Feature customization in your account settings (Grammarly).

Grammarly staff read your text only with your permission, it says. Access needs "review and approval by the responsible managers," and a support team may open your document to fix a problem "with your permission" (Grammarly). Grammarly's privacy page still says its "writing experts will read the content you submit" to the expert writing service (Grammarly), though that service was discontinued on January 31, 2025 (Grammarly).

Governments can ask. Superhuman's policy says it may disclose information "to comply with any applicable law, regulation, legal process, or appropriate governmental request" (Superhuman). The policy doesn't say whether you'd be told first.

How long does Grammarly keep your writing?

Mostly for periods it doesn't name, apart from the few in the table below (Superhuman).

Every retention period Grammarly publishes

Read November 9, 2026.

The license you grant lasts longer than any of these. Grammarly's terms, effective October 29, 2025, give it a license to your content that "lasts as long as intellectual property laws protect your User Content," for uses including improving its services (Grammarly). The terms list "Ownership and License of User Content" among the sections that survive when your account ends.

How do you delete your Grammarly data?

Delete your account. For a free account, sign in, go to the Profile page, and click Delete Account at the bottom (Grammarly). With a paid subscription still active, you submit a support ticket instead. If you use Superhuman's suite, Grammarly says to email care@superhuman.com. On an organization's subscription, the admin has to remove you first.

Grammarly says deleting the account removes your personal information, including saved Editor documents, and "cannot be undone" (Grammarly). To see what it holds before you go, open your privacy settings and click Submit Request next to Personal Data Report, which Grammarly "typically processes" within a few hours (Grammarly).

Has Grammarly been hacked?

We found one publicly reported security bug, from 2018, which its finder rated high severity. Tavis Ormandy of Google Project Zero reported on February 2, 2018 that the Chrome extension, then used by about 22 million people, exposed its login tokens to every website, so "any website can login to grammarly.com as you and access all your documents" (Project Zero). Project Zero's tracker lists the bug as fixed. We found no breach notice in Grammarly's help center or on its trust and security pages as of November 9, 2026. For individual subscribers, Grammarly says it will report a data security incident "in accordance with applicable law" (Grammarly).

Its security claims today are specific. Grammarly says Ernst & Young conducts its annual SOC 2 Type 2 examination, and it holds ISO 27001, 27017, 27018, 27701, and 42001 certifications (Grammarly). It encrypts data in transit with TLS 1.2 and at rest with AES-256 (Grammarly). It runs a bug bounty on HackerOne (Grammarly). It also says "We cannot and do not guarantee absolute security" (Grammarly). Your own account can add two-step verification from its Two-step verification page, though Grammarly says that option is "currently unavailable for Superhuman accounts" (Grammarly).

How Morse keeps your drafts off a writing assistant's servers

Each problem above starts with how a writing assistant works. It sits in the fields where you type and sends that text to servers that keep most of it for periods no page names, under a license that outlasts your account. A setting can narrow where Grammarly runs or what it trains on, but no setting keeps the text on your device.

On Morse, you choose the paragraph and paste it into a chat yourself. The chat is saved in your browser, encrypted, and unlocking it takes a key that never leaves your device plus one Morse gives you when you sign in. Morse's servers don't keep your messages or log what you write, and your account holds only sign-in details, plan, balance, and a usage ledger with no content. Morse doesn't train AI on what you write, sell your data, or show ads, and it runs no advertising trackers or third-party analytics.

Pick the mode by how sensitive the paragraph is:

• E2EE, for the contract clause or personal letter no company should read. On Pro, your browser checks the enclave's hardware attestation before anything is sent, then seals the paragraph so only that verified enclave can open it. Morse's servers pass along ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the protection rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models, without web search or image generation.
• Private, for everyday proofreading and rewrites. New chats start here, on open-weight models such as DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing.
• Anonymous, for a second opinion from GPT, Claude, or Gemini. Morse's proxy removes your account, email, and IP address and sends the text under Morse's own account. The provider sees the paragraph without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so take out names first.

Clearing your browser, losing the device, or deleting your account deletes the chats for good, and Morse can't recover them, so keep the finished draft in your own files. Because we don't store chats, we usually have very little to give authorities. The Morse privacy page shows what each mode lets anyone see, and private AI for writers covers longer manuscripts.

So is Grammarly safe enough for what you write?

For a shopping list or a casual email, yes. Before you check anything more:

• Training: switch off Product Improvement and Training, and if you're on a team account bought on the website, ask your admin whether they have. On Morse, there's no switch to find, because it doesn't train on what you write.
• Where it runs: turn Grammarly off on your bank, your health portal, your work email, and any site you'd rather it didn't read. On Morse, the model gets only what you paste into a chat.
• Authorship: leave tracking off unless an instructor asks for a report, and delete shared reports when the course ends.
• Retention: request a personal data report to see what Grammarly holds, and delete the account when you're done with it. On Morse, your history sits in your browser, encrypted.
• A draft no one else should read: paste it into an E2EE chat on Morse instead.

Paste a paragraph into a chat from the Morse home page without an account, or start free with no card. E2EE chats come with Pro, at $19 a month.