Is Wispr Flow private? Its training default, cloud, and retention

The Morse teamPublished November 2, 20269 min read

As of October 2026, Wispr Flow is not private from Wispr by default: every dictation is sent to Wispr's cloud to be transcribed, and its help center says the setting that lets Wispr train on your audio, transcripts, and edits is on by default for the Free and Pro plans. Wispr's privacy policy, last updated August 19, 2026, still describes that training as something you "opt to share," so Wispr's own pages disagree about the default.

We read Wispr's privacy policy, Data Controls page, privacy and security page, pricing page, Trust Center, and three of its help center articles on October 24, 2026. Every fact below links to the Wispr page it came from. A privacy tracker run by Voibe, which sells a rival dictation app, flagged the training conflict on October 13, 2026, and we checked it against Wispr's pages ourselves. Wispr deserves credit for how much it publishes: its Trust Center names the AI companies that handle your dictation and says which feature sends text to each one.

Wispr Flow's privacy settings at a glance

Does Wispr Flow train on your data?

Yes, on Free and Pro, unless you turn it off. Wispr's security and privacy overview says "Improve the model for everyone" "is on by default for Free and Pro, with the choice shown during onboarding, and you can turn it off" (Wispr). Wispr's setup guide says the same, and lets you pick "Don't use my data for training" on the "You control your data" screen during setup (Wispr).

The setting covers what you say and how you fix it. Wispr's Data Controls page says your "audio, transcript, edits" may be used "to evaluate, train, or improve AI models" (Wispr). The help center adds that the setting applies to Notetaker meetings as well as dictation.

To turn it off, open Wispr Flow, go to Settings, then Data and Privacy, and switch off "Improve the model for everyone" (Wispr). The switch saves at once with no confirmation. Wispr says turning it off "does not stop cloud processing," and Flow still uploads details of each dictation, including "word count, duration, app, URL, timestamp," and your microphone.

Two actions send content to Wispr whatever the switch says. A dictation you report "is retained by Wispr and can be used to train and evaluate its models" (Wispr). A thumbs up or down sends the session too, which "may include audio recordings, transcripts, your edits, the app where you dictated," and authorized staff may review it (Wispr).

For how other AI companies set this default, see which AI companies train on your data.

Where Wispr's own pages disagree

We found five places where one Wispr page says something another contradicts or leaves out, all read on October 24, 2026.

The Data Controls page, last updated August 18, 2026, doesn't state a default for the training switch at all (Wispr). Where the pages disagree, assume the less private answer and check the switch yourself.

What happened to Wispr Flow's Privacy Mode?

Wispr renamed it. The Data Controls page says "Improve the model for everyone" "was previously called 'Privacy Mode,'" and "if you had Privacy Mode enabled, your data is not shared for model training" (Wispr). Privacy Mode on meant training off, so the new switch reads the other way round: off is the private setting.

Wispr's own zero data retention control is for Enterprise. The help center says an "Enforce zero data retention" control can appear in the admin portal, and "only Enterprise admins can change it" (Wispr). On any plan, the Trust Center says that with training and dictation cloud storage both off, "dictation content is not stored on Wispr's servers, except dictations a user chooses to report" (Wispr). For what the term promises in general, see our zero data retention explainer.

What does Wispr Flow capture, and where does it go?

Your voice, plus some of what's on your screen. Wispr's Data Controls page says "transcription always occurs on the cloud," and that Wispr "always uses information about the app you are dictating in" and "the surrounding textbox content" to format and punctuate (Wispr). With Context Awareness on, "relevant text data from the active app window" is used as well. The privacy policy gives "the text on the screen" as an example (Wispr).

The Trust Center names who handles each step, as of October 24, 2026 (Wispr):

• Transcription: Wispr's own speech models, hosted at Baseten and Modal, with Google Cloud routing the traffic. ElevenLabs transcribes languages outside Wispr's main set.
• Formatting and Polish: Wispr's own model, with OpenAI for unsupported languages, as a fallback, and for the Polish feature.
• Command Mode and Transforms: Anthropic and OpenRouter for Command Mode, and Cerebras for Transforms.
• Notetaker: Soniox transcribes meetings, Anthropic writes summaries, and OpenRouter routes meeting content to other model providers.

Wispr's pages say these providers may not train on your data, whatever your setting. The Trust Center adds that Wispr "does not claim zero data retention for Meeting Notetaker" (Wispr).

Your data is stored in the United States. The Trust Center says dictation from Asia is processed in Singapore without being stored there, and Wispr "does not currently offer selectable data residency" (Wispr).

Is Wispr Flow safe?

For security, Wispr publishes its audit details. Its Trust Center says a SOC 2 Type II examination by A-LIGN covered April 23 to July 23, 2026, with "no exceptions noted," and that "the in-scope system is the Wispr Flow dictation service" (Wispr). Wispr also lists an ISO/IEC 27001:2022 certificate. Stored data is encrypted with AES-256 and sent over TLS 1.2 or later (Wispr).

Wispr's help center says plainly that the service "is not end-to-end encrypted in the strict sense, because the service must process audio to produce a transcript," and that Wispr holds the keys (Wispr).

HIPAA needs paperwork. Wispr says you can sign a Business Associate Agreement "on any plan" in Settings, then Data and Privacy, and that accepting one "turns off Notetaker and Scratchpad" (Wispr). A signed BAA locks training off and dictation cloud storage off (Wispr). We cover asking AI about health more broadly in private AI for health questions.

Wispr may disclose personal data to comply with "any court order or legal process or respond to any government or regulatory request" (Wispr). As of October 24, 2026, we found no transparency report from Wispr counting such requests.

How do you delete your Wispr Flow data?

Request deletion in the app, under Settings, then Data and Privacy, and read the confirmation first: Wispr says to "treat deletion as permanent" (Wispr). You can download an export there first, as a ZIP of up to 2 GB that stays available for 30 days.

The Trust Center says deletion removes your account from Wispr's main databases when it runs, clears copies in storage and analytics "within 90 days," and lets backups expire on their 14-day schedule (Wispr). Subprocessors delete data under their own terms. Privacy requests can also go to support+privacy@wisprflow.ai (Wispr).

How Morse keeps what you dictate to AI private

Every setting above decides what happens to a copy of your words on someone else's servers. Wispr's copy is one. Many people dictate straight into an AI chat box, and that app keeps a second copy under its own rules, as our look at whether AI chats are private shows.

Morse is an AI chat app built so the second copy never piles up on a server. Your chat history is stored only in your browser, encrypted, and Morse's servers don't keep messages or log what you write. We don't train AI on your content, sell your data, or show ads (privacy policy). There's no training switch to find, under any name, and no cloud storage setting to recheck after you sign in.

Pick the mode by what you're asking:

• E2EE, for anything you wouldn't want a transcript of. Your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can read it. Morse's servers relay text they can't read. It rests on the enclave hardware and attestation working as designed. Any dictation app that transcribes in the cloud still hears what you say, so type these messages rather than dictating them.
• Private, for everyday questions you'd rather no host keep. It's the default mode. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing, and where a provider confirms that on each reply, Morse checks it and refuses a reply that doesn't.
• Anonymous, for GPT, Claude, and Gemini without your name attached. Our proxy removes your account, email, and IP address before the message goes out under Morse's account. The provider sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy.

Your history is locked when you sign out. Clearing your browser deletes it for good, and Morse can't recover it, so keep a copy of anything you need. Because we don't store chats, we usually have very little to give authorities.

If you keep using Wispr Flow:

• Training: switch off "Improve the model for everyone" in Settings, then Data and Privacy, and don't report dictations or rate them.
• Storage: switch off dictation cloud storage, set local history to auto-delete, and check both again after every sign-in.
• Your screen: turn off Context Awareness if you dictate into apps showing other people's information.
• Meetings: remember Notetaker transcripts are stored in Wispr's cloud with no opt-out, so set a retention period.

You can start free on Morse with no card, and end-to-end encrypted chats come with Pro at $19 a month.