Does Claude save voice recordings? What voice mode and dictation keep

The Morse teamPublished November 2, 20267 min read

As of November 2026, Claude does save a text transcript of every voice mode conversation in your chat history, and Anthropic deletes dictation audio once it has turned your speech into text. Whether Anthropic keeps the audio from voice mode itself, Anthropic's own pages don't say. Since early October 2026, Claude has asked voice users for separate permission to use their "audio recordings and voice chat data" to train its models, a setting reported to be off unless you tap Allow.

We read Anthropic's help pages on voice mode, dictation, and incognito chats, its Privacy Center articles on retention, training, and dictation, and its privacy policy (effective September 10, 2026) on November 2, 2026. The voice-data setting appears on none of them yet, so for that part we rely on reporting from BleepingComputer and Android Headlines and say so where we do. Anthropic deserves credit for making voice training a separate switch, off by default as reported, instead of folding audio into its existing setting for chats.

Claude voice data at a glance

Does Claude keep the audio from voice mode?

Anthropic doesn't say. The voice mode help page answers "Will my voice conversations be saved?" with "Yes," and then only about text: "Textual transcripts of your audio conversations are saved in your chat history just like text conversations" (Anthropic). It says nothing about the recording.

The privacy policy doesn't fill the gap. We searched the full text of Anthropic's privacy policy, effective September 10, 2026, and it never uses the words "voice," "audio," or "speech" (Anthropic). It covers what you say under "Inputs," which it defines by format: "chat, coding, agentic sessions," and "connected services."

The new setting's wording suggests recordings exist, at least for sessions you approve. As reported, it reads: "For approved sessions, allow the use of your audio recordings and voice chat data to improve Anthropic AI models" (BleepingComputer, October 4, 2026). Neither the prompt nor the setting says how long approved recordings are kept, Android Headlines reported (Android Headlines, October 5, 2026).

What is Claude's "Allow us to use your voice data" setting?

It's a training switch for voice, separate from the one for text. Here is what's been reported, since Anthropic hasn't published a help page on it as of November 2, 2026:

• The prompt. Starting a voice conversation shows "Allow us to use your voice data to improve our AI models," with the buttons Allow and Not now (BleepingComputer).
• The reason given. The prompt says "Audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech."
• Where it lives. Settings, then Privacy, under "Allow us to use your voice data," off by default in the version reporters saw.
• What it doesn't touch. It's independent of the text setting, so you can share voice and keep text out, or the reverse (Android Headlines).
• Undoing it. Reports say you can turn it off later and delete the voice data you shared from the same Privacy settings.
• Who sees it. Anthropic hasn't said whether every account gets the prompt or only a test group.

Does Claude train on your voice?

Only if you say yes to the voice setting, but Anthropic's older pages say it never does. Anthropic's Privacy Center article on dictation, dated March 16, 2026, says "We do not use your voice for training our models" (Anthropic). The dictation help page goes further: "We will not retain a copy of your speech input or use it for training our generative models" (Anthropic).

Both sentences sit on pages about dictation, and the October setting asks for voice mode audio. A guide that quotes the dictation line as Claude's answer on voice training is quoting a page about a different feature. As of November 2, 2026, Anthropic hasn't updated either page or added one for voice mode.

Your words can be trained on even with the voice switch off. The dictation article says that if you've allowed Anthropic to use your chats, "the transcribed text may be used in accordance with your privacy settings" (Anthropic). Voice mode transcripts sit in your chat history like typed chats, so the same text setting governs them. Training data includes "the entire related conversation," Anthropic says (Anthropic).

How do you turn off Claude voice training?

Check two switches, because voice and text are separate.

1. Text, including voice transcripts. On the web or desktop, select your name, choose Settings, then Privacy, and toggle off Help Improve our AI models (Anthropic). The phone app has the same path.
2. Voice audio. In the same Privacy screen, leave "Allow us to use your voice data" off, as reported. If a voice chat asks, tap Not now.

Turning the text setting off doesn't pull your chats out of models already trained, as we explain in is Claude private.

Is Claude dictation more private than voice mode?

Yes, on what Anthropic has published. Dictation is the microphone in the Claude mobile app's text box, and it turns your speech into a written prompt, and Claude replies in text (Anthropic). Anthropic deletes the audio after converting it.

Voice mode is a two-way spoken conversation, and in hands-free mode, its default, "Claude listens continuously and responds to natural pauses in your speech" (Anthropic). It can also use the tools you've connected, like Gmail, Google Calendar, Google Docs, and Slack, so a spoken question can pull your email into the chat. Anthropic says raw content from connectors isn't used for training unless it's copied into the conversation (Anthropic).

How long does Claude keep voice chat transcripts?

As long as any typed Claude chat, because the transcript is the chat. The same retention periods apply, which we list in is Claude private, and deleting the chat is how you remove a transcript, step by step in how to delete Claude chats.

Can you use Claude by voice without it being saved?

Not fully. Incognito chats aren't saved to your history, aren't used for training, and are kept for 30 days (Anthropic). Anthropic's incognito page doesn't mention voice mode or dictation, so it doesn't say whether either works there.

Team and Enterprise accounts follow different rules. Anthropic's commercial dictation page says it deletes the audio and doesn't use your voice for training (Anthropic). An Enterprise owner who wants voice mode off for the whole organization has to contact Anthropic's support, the voice mode page says.

How Morse keeps what you'd tell Claude off an account

Every path above ends with what you said stored as text in a chat on Anthropic's servers, under an account that needs a verified phone number. Anthropic says it requires phone verification "for all new users, and there isn't a way to skip this step" (Anthropic). On Morse, a typed chat stays off any account history, and the provider behind it doesn't learn who you are.

• Anonymous, to ask Claude without Anthropic knowing it's you. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account, so no Anthropic account of yours and no phone number is behind it. Anthropic sees the words and may keep them for a limited time to check for abuse, under its own policy, so leave out names and details that would identify you.
• Private, the default, for things you'd rather no host kept. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok. Morse has zero data retention agreements with its inference, search, and other providers, so the hosts agreed to keep nothing.
• E2EE, for what you'd only say to yourself. 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 ciphertext they can't read, and the guarantee rests on the enclave hardware and attestation working as designed. Here's how end-to-end encrypted AI chat works.

In every mode, your chat history is stored in your browser, encrypted, and locked when you sign out. Morse's servers don't keep your messages, and Morse doesn't train AI on your content, sell your data, or show ads, so there's no training switch to check. Because we don't store chats, we usually have very little to give authorities. Clearing your browser deletes your chats for good, and Morse can't recover them, so copy out anything you'll need.

What to do about Claude voice recordings, by worry

• Your voice being trained on: leave "Allow us to use your voice data" off and tap Not now when asked. Morse doesn't train AI on what you write.
• Your words being trained on: turn off Help Improve our AI models, and skip the thumbs up and down on anything personal.
• Stored audio: use dictation rather than voice mode, since Anthropic says it deletes dictation audio.
• A record tied to your phone number: see whether Claude is private, or ask Claude through Morse's Anonymous mode.
• The same questions about ChatGPT: read what ChatGPT keeps from Voice and Dictation.

Try a chat from the Morse home page without an account, or start free with no card needed. End-to-end encrypted chats come with Pro, at $19 a month.