Anthropic Asks Claude Users To Share Voice Data For AI Model Training

Anthropic has started asking Claude users to voluntarily share audio recordings and voice-chat data to help improve its AI models, introducing a separate privacy control specifically for voice interactions.

The reported setting is optional and was disabled by default in the interface examined, allowing users to keep voice data out of training even if they separately permit Anthropic to use text chats or Claude Code sessions.

 

TL;DR

 

• Anthropic is asking Claude users for permission to use voice recordings and voice-chat data for AI model training.
• The reported setting is separate from Claude’s existing chat and coding-session training control.
• Users can choose “Allow” or “Not now,” and the setting was off by default in the interface examined.
• Anthropic’s existing dictation documentation still says dictation audio is deleted and not used for model training.

 

Anthropic Adds A Separate Claude Voice Training Setting

 

Anthropic has begun presenting some Claude users with a prompt asking them to allow their voice data to be used to improve its AI models, according to BleepingComputer.

The prompt states that audio recordings and voice-chat data can help Anthropic’s AI models improve how they understand and respond to speech. Users are offered two choices, “Allow” and “Not now.”

A separate option also appears under Settings > Privacy, labeled “Allow us to use your voice data.”

“For approved sessions, allow the use of your audio recordings and voice chat data to improve Anthropic AI models,” the setting reportedly explains.

Importantly, BleepingComputer found the toggle disabled by default in the interface it examined. MacObserver also reported that the new control gives users a separate choice over whether spoken interactions can be used for training.

Users can reportedly disable the setting later or delete their voice data through Claude’s settings.

 

Voice Data And Chat Training Are Separate

 

The new voice control is separate from Anthropic’s existing setting governing whether regular Claude chats and coding sessions can be used to improve its models.

That means a user could permit voice-data training while excluding their text conversations and Claude Code activity, or allow text-based training while keeping voice recordings excluded.

Anthropic’s consumer model-training documentation says chats and coding sessions from consumer products may be used to improve Claude when users choose to allow it, as well as under limited circumstances such as safety reviews or other explicit opt-in programs.

The company previously made this consent model a central part of its consumer privacy policy. In its 2025 Consumer Terms and Privacy Policy update, Anthropic said Free, Pro and Max users could decide whether their chats and coding sessions would be used for model improvement and could change the setting later.

Anthropic also said data covered by that existing training option could be retained for up to five years when users opt in.

However, the newly reported voice prompt does not establish whether that same retention period applies to audio recordings. Anthropic has not publicly detailed a voice-specific retention period in the materials reviewed for this story.

Claude Dictation Has Different Privacy Rules

 

There is another important distinction for Claude users: voice mode and dictation are not the same feature.

Anthropic’s Claude Voice Mode documentation describes voice mode as a full two-way spoken conversation in which a user talks to Claude and hears spoken responses. Textual transcripts of those voice conversations are saved in chat history.

Dictation, meanwhile, simply converts a user’s spoken prompt into text before Claude responds in writing.

Anthropic’s current Privacy Center guidance for dictation says audio recordings are deleted after speech is converted into text.

“We do not use your voice for training our models,” Anthropic states in the dictation documentation.

The resulting text can still be used for training if a consumer has enabled the separate chat-training setting.

This means Anthropic’s existing statement about not training on voice recordings should not automatically be interpreted as applying to the newly reported voice-data opt-in. The published privacy page specifically describes Claude’s dictation function, while voice mode is documented as a separate feature.

 

What Remains Unclear About Claude Voice Data

 

The new consent prompt provides users with an explicit choice over whether to share voice information, but some details remain unanswered.

Neither BleepingComputer’s report nor Anthropic’s currently available voice and privacy documentation specifies how long voice recordings permitted for training would be retained, whether the feature has reached every Claude user, or whether availability differs by region or subscription.

The reported wording also refers to “approved sessions,” but Anthropic has not publicly explained in detail what qualifies a session for inclusion.

For users who want to keep their spoken conversations out of model training, the reported control can be checked under Claude’s Privacy settings.

The key distinction is that sharing voice data is not described as mandatory. Based on the interface examined so far, Anthropic is asking users to opt in rather than automatically enrolling them in voice-data training.