Anthropic asks Claude users to share voice for AI training

Anthropic has begun asking Claude users whether it can use their voice conversations to train and improve its AI models. The request is optional, and the setting behind it is off by default.

The prompt shows up when people use Claude's voice features. It asks directly for permission to use audio recordings and voice chat data for model training.

"Allow us to use your voice data to improve our AI models," the prompt reads. It goes on to explain: "Audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech. Turn this off or delete this data anytime in settings."

Users get two choices in the prompt: Allow or Not now. Neither one is forced, and the decision can be changed later.

A separate switch in the privacy settings

Voice training is not part of the existing toggle that controls training on regular chats and coding sessions. Anthropic has added a dedicated option for it under Settings > Privacy, labelled "Allow us to use your voice data."

The description of that setting reads: "For approved sessions, allow the use of your audio recordings and voice chat data to improve Anthropic AI models."

Because the two controls are independent, users can mix and match. Someone could let Anthropic train on their spoken conversations while keeping text chats and Claude Code sessions out of the training pool. Someone else could do the opposite and share chats and coding sessions but keep their voice recordings private.

In the version of the interface seen so far, the voice setting is disabled by default. Anthropic is therefore not quietly opting users in. Training on voice data only happens if a user actively agrees, either through the pop-up prompt or by flipping the toggle in settings.

Turning it off and deleting data

The company says users can switch the option off at any point. It also says voice data can be deleted from the settings menu if someone changes their mind after agreeing.

The prompt itself points to this, telling users they can "turn this off or delete this data anytime in settings." The phrase "for approved sessions" in the setting description suggests that consent applies to sessions where the option is enabled, rather than to past voice chats recorded before a user agreed. The source material does not explain in more detail how this works in practice.

Anthropic has not given further details about how long voice recordings are kept or how they are processed once a user allows training.

Why the separation matters

Splitting voice from text gives users finer control over what leaves their hands. Voice recordings are a different kind of data from typed messages. A recording carries the sound of a person's voice along with whatever they say, and spoken conversations can capture background noise and other people nearby.

For users who already allow training on their chats, the new toggle means they need to make a second, separate decision about audio. For those who have opted out of chat training, nothing changes for their voice data unless they choose to allow it.

Our Take

The default-off setting and the separate toggle are the parts that matter most here. Many AI services have been criticised for burying training consent inside a single broad switch. A dedicated control for audio suggests Anthropic sees voice as more sensitive than text and wants consent to be specific.

The move fits a wider pattern of AI vendors seeking more training data while facing closer attention from regulators. Anthropic and OpenAI are already the subject of an FTC inquiry into consumer risks, and how companies collect and use personal data is likely to be part of that discussion. Anthropic has also recently expanded how Claude data flows to enterprise tools through its Compliance API, which shows how many paths user content can now take.

Readers who use Claude's voice features should check the Privacy settings rather than relying on the pop-up alone. It is worth watching whether the option stays off by default over time, whether Anthropic publishes retention details for audio, and whether other AI vendors follow with separate consent controls for voice data.